How can we accurately predict for matter of taste using opinions from dissimilar individuals?

How can we accurately predict for matter of taste using opinions from dissimilar individuals? | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article How can we accurately predict for matter of taste using opinions from dissimilar individuals? Itsuki Fujisaki, Kunhao Yang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7636367/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract People frequently rely on the opinions of others to make decisions regarding matters of taste, such as choosing movies or restaurants. Then, how can we get accurate prediction from others' opinions? In this respect, similarity and majority-based strategies have been studied. However, the potential benefits of relying on dissimilar individuals' opinions remain unclear. This study investigates the efficacy of a novel strategy, termed the "Dissenting preference strategy". This strategy involves making a different (or opposite) choice from that of a person whose preferences differ from one's own. We used computer simulations based on the dataset comprising ratings from 14,000 individuals. Our results reveal that this strategy can improve decision-making accuracy, particularly when a moderate number of experiences (25 or more) and a certain neighborhood size (around 20 individuals or more) are available. However, the similarity-based strategy ("Doppelgänger strategy") consistently outperformed the Dissenting preference strategy. Nevertheless, further analyses highlighted that individuals with lower mean taste similarity could benefit more from the Dissenting preference strategy than from Doppelgänger strategy. Additionally, incorporating multiple dissimilar individuals (up to four) slightly enhanced the accuracy of the strategy. These findings underscore the conditional utility of dissimilar people in decision-making and suggest avenues for integrating dissimilarity into recommendation systems, especially for users with atypical preferences. Physical sciences/Mathematics and computing Biological sciences/Psychology Social science/Psychology Social learning Online recommendation Matter of taste Dissimilarity Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Which movie should I watch next? Where should I go for today’s dinner? We make decisions about matters of taste on a daily basis, on which there are no universal, objective criteria. In these cases, we routinely consider the opinions of others who already experienced and evaluated the options. Then, can we make better decisions to consider them? Studies in social learning, advice-taking etc., 1–9 have examined this issue and proposed a formal model of social learning. However, these studies mainly investigated matters of fact which has objective criterion. In other words, they did not consider matters of taste, for which there are no objective criteria. Importantly, some recent studies 10 – 15 have addressed the matter of taste. Especially a significant study by Analytis et al.(2018) 10 proposed several models of social learning strategies based on recommender system algorithms and those about matters of fact and when and how these different strategies thrive or fail. Figure 1 illustrates the case where people will rely on others’ opinions. Imagine that you have already experienced three albums and are thinking about which album to buy or stream. Now, you can use the opinions of four people who have also experienced the albums. Moreover, they show the preferences of the two albums you are considering. Then, how should we use others’opinions? Simply put, Analytis et al.(2018) 10 proposed two strategies: ①Similarity: through experiencing some albums, you can identify who has a similar tastes to you. In Fig. 1, X_3 has almost the same taste as you. Therefore, you will choose \(\:\varvec{T}\_2\) because s/he preferred T_2 to T_1 (hereafter, called “Doppelgänger”). ②Majority: You can also rely on the opinions of the majority, as with the wisdom of crowds. In this case, you will choose \(\:\varvec{T}\_1\) since three of the four others preferred T_1 to T_2. Analytis et al.(2018) 10 reported that they could work effectively. In other words, these strategies could predict your future satisfaction above the chance level. However, in daily life, we often can not get similar other’s opinions and many others’ opinions. Can we make better decisions even in this case? In this respect, descriptive studies in social and consumer psychology have pointed out the importance of receiving the opinion of “dissimilar” others. Although some studies show that the opinion of dissimilar others is ignored or discounted, others have indicated that in forming opinions, people take irrelevant cues into account to the extent 16 , 17 . In particular, recent studies showed that the opinions of dissimilar people made a person make contrasting choices. For example, when there are items A and B, and the other you dislike, choose item A; you will like the contrast item. In other words, you will choose item B. In addition, some studies have shown that contrast choice emerged as a cognitive process (i.e., not by disliking). Tuk et al.(2019) 18 demonstrated that perceptions of dissimilarity trigger a more general hypothesis of dissimilarity. In the experiment (Ex. 1-a), participants read texts about other who had different sex, different taste of jokes, and did different judgmental decision from the participants. Then, they read additional texts, and the other recommended one of the two experimental tasks. Then, the participants were required to answer which tasks they wanted to do. The results showed that the participants chose the task the others did not recommend. Let’s go back to our illustration. In Fig. 1, X_2 has a highly dissimilar taste to you. Therefore, based on the aforementioned studies, you seem to choose T_2 against the recommendation of X_2. However, it remains unclear whether this strategy (hereafter called “Dissenting preference strategy”) enables people to make better decisions since the previous research only empirically observed people’s behaviour. Using a series of computer simulations on a large-scale dataset, we investigate this issue. As a dataset, we used the Jester dataset 10 , which consists of 14,000 people’s 100 joke ratings. Through the simulations, we first identified whether Dissenting preference strategy enables participants to make better decisions. In particular, we point to the significant role of the number of experiences and the size of the neighbourhood. In addition, we compared the Dissenting x preference strategy with the other strategy based on the similarity (Doppelgänger) and found that the Doppelgänger strategy usually beat the Dissenting preference strategy. However, we also analysed individual differences to identify those who can leverage the efficacy of the Dissenting preference strategy more than the Doppelgänger strategy. -----Figure 1 about here----- Results Analysis Based on Analytis et al.(2018) 10 , we constructed the simulation architecture. Figure 2 shows the one trial of the simulations. (1) We select 250 group members from all the people (= 14,000). (2) We randomly allocate jokes to either training or test trials. (3) We select the targeted person (“You”). (4) We pick up the person who has the most dissimilar taste compared with You of all group members (“Dissimilar Person”). We conduct this procedure using the Pearson correlation in training trials (5, 10, 15...75 trials). (5) You and the Dissimilar Person experienced test trials. The test trials consist of binary choice problems from all jokes in test trials. We created 300 binary choice problems by combination (= 25 * 24 / 2). We test whether the Dissenting preference strategy can predict Your future satisfaction in the test trials. Note that the Dissenting preference strategy predicts that You will like the joke that the Dissimilar Person does not like from the binary choice problem. We repeated each step in this procedure, and We ran the simulations 1,000 times for each number of training trials (5, 10, 15...75 trials). Note that in one trial of the simulations, we tested all 250 people as You. In other words, we repeated (3) ~ (6) for 250 times. -----Figure 2 about here----- Main results Notably, in the simulations, we set the number of group members at 250, as in Analytis et al.(2018) 10 . The assumption is based on the arguments on Dunbar’s number 19 . The literature proposed that people can build stable relationships up to around 250 because of the cognitive limit. Therefore, the simulations treat the case where You seek opinions around the community, which You can access cognitively. Figure 3a shows the results of the analysis. This figure indicates the relationship between the mean accuracy and the number of training trials. We can see that the larger the number of trials, the higher the mean accuracy. Especially from 20 to 25 trials, the mean accuracy exceeded the chance level. Therefore, dissimilarity enabled us to make better decisions, especially above 25 trials. Conversely, to make a good decision, dissimilarity (i.e., Dissenting preference strategy) requires people to do training even 25 trials. Figure 3b adds the result of the additional simulations. In particular, the figure shows the results of the Doppelgänger strategy, as in Analytis et al.(2018) 10 . In the additional simulations, the Doppelgänger strategy chose the person most similar to You in the training trials. In the test trials, the Doppelgänger strategy predicted that You would like the joke that the most similar person did like from the binary choice problem. As with the Dissenting preference strategy, the larger the number of trials, the higher the mean accuracy in the Doppelgänger strategy. Notably, the Doppelgänger strategy recorded higher accuracy than the Dissenting preference strategy in any number of trials. Moreover, the Doppelgänger strategy beat the Dissenting preference strategy when the number of trials in the Doppelgänger strategy was minimum (= 5) and that in the Dissenting preference strategy was maximum (= 100). In addition, contrary to the Dissenting preference strategy, the Doppelgänger strategy recorded above the chance level at any number of trials. Then, how did the results emerge? Fig. 3c indicates the results of further analysis. The analysis examined the relationship between the correlation in the test trials and the number of training trials for the Dissenting preference. The correlation was between You and a Dissimilar Person. First, in Dissenting preference strategy, the larger the number of trials, the lower the correlation. The results indicated that many trials enabled You to pick up a fully dissimilar person. Notably, the correlation became below zero when the number of trials was 25. Thus, we confirmed that the value of the correlation determined the accuracy of the Dissenting preference strategy. As explained in the Analysis section, the Dissenting preference strategy predicts that You will like the joke that a Dissimilar Person does not like. Overall, we found that people could make good decisions based of dissimilarity, but dissimilarity’s effectiveness was relatively low and limited. We shall make a note of this issue in the Discussion section. -----Figure 3 about here----- Individual differences Hereafter, we investigated individual differences in Dissenting preference strategy. In particular, we aimed to identify those who can leverage the Dissenting preference strategy more than the Doppelgänger strategy. In this respect, Analytis et al.(2018) 10 proposed two factors that play a role in individual differences: mean taste similarity across all other individuals (hereafter called “Mean taste similarity”) and variance across all individuals in taste similarity (“Dispersion in taste similarity”). For example, Analytis et al.(2018) 10 demonstrated that the accuracy of the Doppelgänger strategy was high for those with high Mean taste similarity and high Dispersion in taste similarity. Based on Analytis et al.(2018) 10 , we first calculated Mean taste similarity and Dispersion in taste similarity with all other people (= 13,999 people) and then conducted the simulations. We conducted the simulations by regarding each individual as You. In this simulation, we set the number of trials at a maximum (= 75). Except for these assumptions, the simulations were the same as the main simulations. Figure 4a shows the results of the simulations in Dissenting preference strategy. The mean accuracy tended to be high in large Dispersion in taste similarity. When the Dispersion in taste similarity is large, You can pick up the person with a dissimilar taste. This tendency was mitigated by high Mean taste similarity. The mean accuracy tended to be especially low in low Mean taste similarity. This is because when Mean taste similarity is high, it is relatively hard for You to find the dissimilar person. Figure 4b shows the results of the simulations in the Doppelgänger strategy. The Doppelgänger strategy seeks a person who has a similar taste to You. Therefore, contrary to the Dissenting preference strategy, Mean taste similarity should be high to achieve high accuracy. Next, similar to the Dissenting preference strategy, Dispersion in taste similarity should also be high. This is because, in this case, there seems to be a person with an exceptionally similar taste to You. In Fig. 4c, we compared the Dissenting preference strategy with the Doppelgänger strategy. Especially for each individual, we subtracted the mean accuracy in the Doppelgänger strategy from the mean accuracy in the Dissenting preference strategy. The value was above zero for individuals who recorded higher accuracy in the Dissenting preference strategy than in the Doppelgänger strategy. Although the results were ambiguous for Dispersion in taste similarity, we can see a clear tendency for Mean taste similarity. When Mean taste similarity was low (e.g., -0.1), some points were represented in orange to red. For the individuals, the Dissenting preference strategy recorded higher accuracy than the Doppelgänger strategy. The results are derived from the following findings. As mentioned above, in the Dissenting preference strategy, the mean accuracy tended to be high with low Mean taste similarity. On the contrary, in the Doppelgänger strategy, the mean accuracy tended to be low in low Mean taste similarity. Therefore, in low Mean taste similarity, the Dissenting preference strategy could beat the Doppelgänger strategy for some individuals. In summary, we found conditions under which the Dissenting preference strategy could be highly accurate. Moreover, through the comparison, we identified situations under which the Dissenting preference strategy could record higher accuracy than the Doppelgänger strategy (i.e., Mean Taste Similarity was low). -----Figure 4 about here----- Additional analysis 1: Group size In the main simulations, we set the group size where an individual could seek opinions at 250. This assumption was based on Dunbar’s number 19 . However, we do not get the exact number of people an individual can rely on. Therefore, it is worth manipulating the group size and examining the effectiveness of the Dissenting preference strategy. Here, we set group sizes at 4, 10, 20, 50, 100, 250, 500, 1000. For simplicity, we set the number of trials at 75. Except for the assumptions, we run the simulations as with the main simulations. Figure 5 shows the results of the simulations. This figure indicates the relationship between the mean accuracy and the group size. We can see that the larger the group sizes, the higher the mean accuracy. Especially from the size of 10 to 20, the mean accuracy became above the chance level. Therefore, dissimilarity enabled us to make good decisions, especially above 20. In anthropology, a band society 20 was proposed as the simplest form of human society. The band generally consists of an extended family or clan. The size of the band is estimated at 30. Therefore, the results implied that dissimilarity enabled us to make good decisions in the band society. This figure also shows the result of the Doppelgänger strategy. As the figure shows, the larger the group sizes, the higher the mean accuracy. In addition, the Doppelgänger strategy was recorded above the chance level at any number of trials, contrary to the Dissenting preference strategy. Moreover, the Doppelgänger strategy recorded higher accuracy than the Dissenting preference strategy in any group size. Importantly, the Doppelgänger strategy beat the Dissenting preference strategy when the group size in the Doppelgänger strategy was minimum (= 4) and that in the Dissenting preference strategy was maximum (= 1,000). Thus, we found the effectiveness and the limitation of dissimilarity for the group size. -----Figure 5 about here----- Additional analysis 2: Dissimilar people How can we enhance the effectiveness of the Dissenting preference strategy? Analytis et al.(2018) 10 indicated that increasing the number of people you rely on can enhance the prediction strategy. Especially by taking 10 of the most similar people, Analytis et al.(2018) 10 showed that prediction accuracy improved by more than 5%. Based on the findings, we conducted additional simulations that took Dissimilar people beyond one person. In particular, we set the number of most Dissimilar people at 1 (i.e., same as main simulations), 2, 3, 4, 6, 10, 20, 50, and 100. Note that preference prediction was conducted using the majority rule 6 , 10 . For example, when Dissimilar people were 6, and four people liked A and two people liked B, the judge by the Dissimilar people was A. therefore, the preference prediction was correct when the target liked B more than A. If a half of Dissimilar people liked A and another half of Dissimilar people liked B, the strategy predicted the preference randomly. Figure 6a shows the results of the additional simulations. This figure indicates the relationship between the mean accuracy and the number of Dissimilar people. As the figure displays, the line of the mean accuracy is concave down: The mean accuracy increased up to three with group size, though slightly and then it decreased. In other words, we could identify an optimal number of Dissimilar people. This figure also shows the results of the Doppelgänger strategy. We can also see that the Doppelgänger strategy's mean accuracies were inverted-U shape. The mean accuracy increased to fifty with group size, though slightly, then decreased. Then, how did the results emerge? We especially focused on the findings that the mean accuracies were inverted-U shape for Dissimilar people in the Dissenting preference strategy. Figure 6b displays the results of the analysis. This analysis calculated the mean accuracy in all group members and then sorted them according to the rank of dissimilarity. We can see that the relationship between the rank and the mean accuracy was inversely proportional. Particularly in Rank 6, the accuracy was below the chance level. Therefore, adding a small number of members could enhance the accuracy; however, adding a large number of people decreased the accuracy. -----Figure 6 about here----- Discussion In everyday decisions, we routinely consider the opinions of others who already experienced them. It is known that relying on similar others or many others enables us to make better decisions. However, we often cannot rely on them. In this case, how can we leverage other’s opinions? Based on the behavioural studies, we focused on the strategy of dissimilarity: One should choose a different opinion from a dissimilar person chose. Especially in the case where there are binary options A and B, and a dissimilar person chose A, one should choose B. To test the strategy's performance, we conducted computer simulations. In the simulations, we picked an individual who was ready to choose binary options and another person who had a dissimilar taste to the individual. Then, we tested the strategy's performance. The results showed that the strategy of dissimilarity could accurately predict the individual’s future satisfaction. Then, we compared the strategy in the previous study: the strategy of similarity. As a result, the strategy of dissimilarity was largely beaten by this strategy. Therefore, we attempted to identify the conditions where the strategy of dissimilarity performed better. As a result, we found that for those who had different tastes from others, the strategy of dissimilarity could be performed better than the strategy of similarity. Moreover, we examined the situation where the strategy of dissimilarity utilised multiple people. As a result, we identified the optimal number for the people (= 4). Through the simulations, we also clarified the conditions under which the strategy of dissimilarity performed well: a moderate number of training trials and a moderate number of members were needed for the strategy to accurately predict one’s future satisfaction. While many previous studies on social learning treated as a matter of fact 1 – 9 , this paper focused on the matter of taste 10 – 13 , for which there is no objective criteria, along with Analytis et al.(2018) 10 . Importantly, to the best of our knowledge, this is the first study that showed “dissimilarity” could lead to better decisions. In other words, our study implied that dissimilarity could become “information” for those who are ready to make decisions. As we mentioned in the Introduction, studies in social and consumer psychology have pointed out the importance of receiving the opinion of “dissimilar” others. Especially, Tuk et al.(2019) 18 showed that we chose different option form the dissimilar person. Combined with the findings in our study, the strategy may be “adaptive” 22,23 . However, it should be noted that our findings——the strategy of dissimilarity could predict the individual’s future satisfaction——have several limitations: a moderate number of training trials and a moderate number of members were needed. Therefore, future studies which examine whether the strategy of dissimilarity performs well directly should be needed. Another promising approach is to test the findings in our study could extend a matter of fact. For example, when an individual sees that another person keeps making mistakes in social learning 1 – 9 , the individual might make accurate decision by making different decision from another person. It is important to examine this point, including how people actually engage in social learning. Moreover, our findings could contribute to real-world review sites. Online review sites have usually utilised recommender systems 23 – 29 (e.g., content-based, demographic-based and personality-based recommender systems). They mainly exploit the users who have similar pattern to the target user. However, this study demonstrated that participants could leverage other users who have dissimilar preference patterns. Therefore, we may build new recommender system based of dissimilarity. Importantly, the recommender system based of dissimilarity could complement usual recommender systems. Analytis et al.(2023) 28 tested representative collaborative filtering algorithms and found the “Mainstream-taste bias”: they perform much worse for non-mainstream or unusual tastes than mainstream users. In this respect, as we mentioned in the Individual Differences section in Results, the strategy of dissimilarity performed well for those with unusual tastes. Therefore, one promising approach is to construct “hybrid” recommender systems. Put simply, by combining collaborative filtering with it, we might enhance the quality of the recommender system for various types of people. One limitation of this study is the architecture of the simulation. In the test trials, a target and another person (people) made only a binary choice. On the contrary, in the real world, we choose an option from a wide range of options. Therefore, it is important to test the strategy for dissimilarity in a wide range of choice situations. In the future, we will re-conduct simulations when there are more options. Another limitation was as for stimuli: This study focused on joke, as with the previous study 10 . However, it remains unclear whether the findings of this study could apply to the everyday’s consumer behaviour, as illustrated in Introduction. Therefore, future studies which examine another range of data should be conducted Methods Dataset As with Analytis et al.(2018) 10 , we utilised The Jester dataset for a large-scale dataset. In this site ( http://eigentaste.berkeley.edu .), people voluntarily read and rate jokes on a scale ranging from not funny (− 10) to funny (+ 10). In other words, people rated the jokes on a scale of 20. Along with Analytis et al.(2018) 10 , again, we only used the data of participants who evaluated all jokes (reducing the number of participants from 73,421 to 14,116). Specifically, we analysed data from which Analytis et al.(2018) 10 randomly selected 14,000 people from the 14,116 people. Analysis of (dis) similarity We used the Pearson correlation coefficient as a measure of similarity between two individuals or two items where (dis)similarity between two individuals i and j is defined as $$\:w(i,\:j)\:=\:\frac{{\sum\:}_{n\:=\:1}^{M}({u}_{in}\:-\:{ū}_{i})({u}_{jn}\:-\:{\text{ū}}_{\text{j}})}{\sqrt{{\sum\:}_{n\:=\:1}^{M}{({u}_{in}\:-\:{\text{ū}}_{\text{i}})}^{2}}\sqrt{{\sum\:}_{n\:=\:1}^{M}{({u}_{jn}\:-\:{\text{ū}}_{\text{j}})}^{2}}}$$ For the similar options strategy, (dis) similarity is calculated using the Pearson correlation coefficient but between two items k and l . M stands for the total number of options or people and n for specific options or people. We chose to adopt the Pearson correlation because in the recommender system literature, it is the canonical measure. Declarations DECLARATION OF INTERESTS The authors declare no competing interests. FUNDING This research received no external funding. We used only open data and simulation analysis. Author Contribution IF and YK developed the study concept and contributed to the study design. IF conduct the simulations and wrote the manuscript, with feedback from KY. Acknowledgement We thank Kazuhiro Ueda for their helpful comments. Data Availability The R-code and the datasets analysed during the current study are available in the Mendeley Data: https://data.mendeley.com/datasets/6s46dfytvp/1 References Boyd, R. & Richerson, P. J. Culture and the Evolutionary Process (Univ. Chicago, 1985). Clemen, R. T. Combining forecasts: a review and annotated bibliography. Int. J. Forecast. 5 , 559–583 (1989). Kameda, T. & Nakanishi, D. Cost–benefit analysis of social/cultural learning in a nonstationary uncertain environment: an evolutionary simulation and an experiment with human subjects. Evol. Hum. Behav. 23 , 373–393 (2002). Kameda, T. & Nakanishi, D. Does social/cultural learning increase human adaptability? Rogers’s question revisited. Evol. Hum. Behav. 24 , 242–260 (2003). Laland, K. N. Social learning strategies. Anim. Learn. Behav. 32 , 4–14 (2004). Hastie, R. & Kameda, T. The robust beauty of majority rules in group decisions. Psychol. Rev. 112 , 494–508 (2005). Larrick, R. P., Mannes, A. E. & Soll, J. B. in Frontiers in Social Psychology:Social Judgment and Decision Making (ed. Krueger, J. I.) 227–242Psychology Press, New York, NY, (2012). Goldstone, R. L., Wisdom, T. N., Roberts, M. E. & Frey, S. Learning along with others. Psychol. Learn. Motiv . 58 , 1–45 (2013). Witt, A., Toyokawa, W., Lala, K. N., Gaissmaier, W. & Wu, C. M. Humans flexibly integrate social information despite interindividual differences in reward. P Natl. Acad. Sci. 121 , e2404928121 (2024). Analytis, P. P., Barkoczi, D. & Herzog, S. M. Social learning strategies for matters of taste. Nat. Hum. Behav. 2 , 415–424 (2018). Yaniv, I., Choshen-Hillel, S. & Milyavsky, M. Receiving advice on matters of taste: Similarity, majority influence, and taste discrimination. Organ. Behav. Hum. Decis. Process. 115 , 111–120 (2011). Müller-Trede, J., Choshen-Hillel, S., Barneron, M. & Yaniv, I. The wisdom of crowds in matters of Taste. Manage. Sci. 64 , 1779–1803 (2017). Barneron, M., Allalouf, A. & Yaniv, I. Rate it again: Using the wisdom of many to improve performance evaluations. J. Behav. Decis. Mak. 32 , 485–492 (2019). Fujisaki, I., Honda, H. & Ueda, K. A simple cognitive method to improve the prediction of matters of taste by exploiting the within-person wisdom-of-crowd effect. Sci. Rep. 12 , 12413 (2022). Kopsacheilis, O. et al. Crowdsourcing the assessment of wine quality: Vivino ratings, professional critics, and the weather. J. Wine Econ. 19 , 285–304 (2024). Schwarz, N. & Clore, G. L. Mood, misattribution, and judgments of well-being: Informative and directive functions of affective states. J. Pers. Soc. Psychol. 45 , 513–523 (1983). Meyers-Levy, J. & Sternthal, B. A 2-factor explanation of assimilation and contrast effects. J. Market Res. 30 , 359–368 (1993). Tuk, M. A., Verlegh, P. W., Smidts, A. & Wigboldus, D. H. You and I have nothing in common: The role of dissimilarity in interpersonal influence. Organ. Behav. Hum. Decis. Process. 151 , 49–60 (2019). Dunbar, R. How Many Friends Does One Person Need? Dunbar’s Number and Other Evolutionary Quirks (Faber & Faber, 2010). Williams, B. J. & Wobst, H. M. A model of band society 29i–138 (Memoirs of the Society for American Archaeology, 1974). Gigerenzer, G., Todd, P. M., Group, A. B. C. R. & T Simple heuristics that make us smart (Oxford University Press, 2000). Goldstein, D. G. & Gigerenzer, G. Models of ecological rationality: the recognition heuristic. Psychol. Rev. 109 (1), 75–90 (2002). Pazzani, M. & Billsus, D. in in The Adaptive Web . 325–341 (eds Brusilovsky, P., Kobsa, A. & Nejdl, W.) (Springer, 2007). Nunes, M. A. S. & Hu, R. Personality-based recommender systems: an overview. In Proc. 6th ACM Conference on Recommender Systems 5–6ACM, (2012). Bobadilla, J., Ortega, F., Hernando, A. & Gutiérrez, A. Recommender systems survey. Knowl. Based Syst. 46 , 109–132 (2013). Fernández-Tobas, I., Braunhofer, M., Elahi, M., Ricci, F. & Cantador, I. Alleviating the new user problem in collaborative filtering by exploiting personality information. User Model. User Adapt. Interact. 26 , 221–255 (2016). Analytis, P. P., Barkoczi, D., Lorenz-Spreen, P. & Herzog, S. The structure of social influence in recommender networks. In: Proceedings of the web conference 2020, 2655–61. WWW ’20. New York, NY, USA: association for computing machinery (2020). Analytis, P. P. & Hager, P. Collaborative filtering algorithms are prone to mainstream-taste bias. In Proceedings of the 17th ACM Conference on Recommender Systems (pp. 750–756). (2023). Analytis, P. P., Kaushik, K., Herzog, S., Bahrami, B. & Deroy, O. A recommender network perspective on the informational value of critics and crowds. (2024). arXiv preprint arXiv:2403.18868. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7636367","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":516891106,"identity":"9c2e9960-1c43-4774-b0f5-b0d919036616","order_by":0,"name":"Itsuki Fujisaki","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA1UlEQVRIiWNgGAWjYBACAyBm/MDAwMwP4iUUEKmFWQKIJRtAWgyI1MLAA2IcgHMJAHPp7jQJyR217MbnVyd+eGDAIM8vdgC/Fss5ZzcbFJ45zmx24+1mCaDDDGfOTiDgsBu5Gx9Ith0Dajm7AaQlweA2YS0bDvACtRjPOLv5B7FaNj7gbathNuDv3UacLSC/GEu2HWCWuMG7zSLBQIKwX8yle7dJfmyrS+bvP7v55o8KG3l+aQJaGCTA5OFkBokEBJcYLXV2DPwHiFA9CkbBKBgFIxIAANUgRp1w8P0DAAAAAElFTkSuQmCC","orcid":"","institution":"Tsukuba University","correspondingAuthor":true,"prefix":"","firstName":"Itsuki","middleName":"","lastName":"Fujisaki","suffix":""},{"id":516891107,"identity":"766bd849-22e2-48e4-b54b-7db9a8249709","order_by":1,"name":"Kunhao Yang","email":"","orcid":"","institution":"Shibaura Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Kunhao","middleName":"","lastName":"Yang","suffix":""}],"badges":[],"createdAt":"2025-09-17 06:23:26","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7636367/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7636367/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":91718352,"identity":"7d4776a6-0170-4ffa-b597-b82f67ecd918","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":123463,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/50fd49d83c8aff5f0d08ec9f.png"},{"id":91718349,"identity":"943adb19-0c95-4502-81ce-af0e1bab7931","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":49142,"visible":true,"origin":"","legend":"","description":"","filename":"DissimilarMain4.docx","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/f360b12354395a31bfdc5821.docx"},{"id":91718351,"identity":"3a86a9dc-5f7d-4cca-b36f-876d287bbeae","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":253457,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/61110f8fd66ea37be3e25e22.png"},{"id":91718356,"identity":"c93b3aad-6926-4a09-b235-4267ae3945f1","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":228744,"visible":true,"origin":"","legend":"","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/87c1f1e022a437bf5991dd41.png"},{"id":91718871,"identity":"b50b76ea-a2ef-4761-80da-2b38004137de","added_by":"auto","created_at":"2025-09-19 13:50:45","extension":"png","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":879664,"visible":true,"origin":"","legend":"","description":"","filename":"Figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/2098100ba06cde63d18b50d6.png"},{"id":91718367,"identity":"23b3634a-d139-415a-bf86-4104d1aea348","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":140911,"visible":true,"origin":"","legend":"","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/a684046920545a1d5cd02d67.png"},{"id":91718874,"identity":"b07e1a23-ab85-445d-b4c3-f1779d0a191d","added_by":"auto","created_at":"2025-09-19 13:50:45","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":131923,"visible":true,"origin":"","legend":"","description":"","filename":"Figure6.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/d9f4c909bc7d93d5dceabae3.png"},{"id":91718377,"identity":"4fec6c08-de71-461a-996a-9f096a4d4a83","added_by":"auto","created_at":"2025-09-19 13:43:10","extension":"json","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4356,"visible":true,"origin":"","legend":"","description":"","filename":"8c7306d65810469392102dc827410b60.json","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/1612559d39a3409492c012f4.json"},{"id":91718876,"identity":"9057ab9b-7329-4f4c-bcf0-b8fbcf9b4afd","added_by":"auto","created_at":"2025-09-19 13:50:45","extension":"xml","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":75382,"visible":true,"origin":"","legend":"","description":"","filename":"8c7306d65810469392102dc827410b601enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/d8216d822b14de46cd321dab.xml"},{"id":91719847,"identity":"3c32a53b-9ce5-4577-b933-3adccfe0df89","added_by":"auto","created_at":"2025-09-19 13:58:45","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":123463,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/efa0ee514ed854630661ad86.png"},{"id":91718371,"identity":"39c168a7-1266-479c-8216-6011acb2d614","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":253457,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/320c17638a9e62d98f936da8.png"},{"id":91718362,"identity":"e376b26a-12d8-4c51-b590-bd65be72e665","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":228744,"visible":true,"origin":"","legend":"","description":"","filename":"Figure3.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/2aa9c81fa0e9c129062cff5d.png"},{"id":91718358,"identity":"32f36ba2-a100-41f1-aad7-29d6b0b41637","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":879664,"visible":true,"origin":"","legend":"","description":"","filename":"Figure4.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/47c70a403e6601bf3886d23f.png"},{"id":91718370,"identity":"c6d529c4-73a8-4b24-98d8-886a26614072","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":140911,"visible":true,"origin":"","legend":"","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/8852a8729235f1bf048010a1.png"},{"id":91718365,"identity":"a1f6ee5a-cd13-4098-8a4b-df01bc6e9798","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":131923,"visible":true,"origin":"","legend":"","description":"","filename":"Figure6.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/d2c3c3f39abe73f4c3d453bc.png"},{"id":91718374,"identity":"88a5151a-0e22-45a4-bebf-dde7655e097d","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":33282,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure1.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/216fd42087c040d42ec52360.png"},{"id":91718878,"identity":"e91351a1-1cde-498a-b3f8-95a1d2a23272","added_by":"auto","created_at":"2025-09-19 13:50:45","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":66145,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure2.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/a5244882767c38f82424c176.png"},{"id":91718361,"identity":"e250f435-078d-44c0-91cd-a9d4fd24667b","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":76550,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure3.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/64c376a9d7ce6a5b121e1b50.png"},{"id":91718879,"identity":"c68539c8-a180-4053-93e2-1937952b0077","added_by":"auto","created_at":"2025-09-19 13:50:45","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":200574,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure4.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/96abc74e4ba4d388daf7c7ef.png"},{"id":91718368,"identity":"10674041-1d58-4ebc-a357-32b5ba309ad5","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":26573,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure5.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/d4cec3bcc1a1249b4566c40c.png"},{"id":91718373,"identity":"a7e46433-6eda-4f71-9a86-8f98f2558b33","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":47310,"visible":true,"origin":"","legend":"","description":"","filename":"OnlineFigure6.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/5fb71cce160295a522bed9ac.png"},{"id":91718375,"identity":"9c0734bf-3aca-41a9-be2c-4cf3ac78f9b7","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"xml","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":73980,"visible":true,"origin":"","legend":"","description":"","filename":"8c7306d65810469392102dc827410b601structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/44bc5f9dc1b02b7350a8de18.xml"},{"id":91718880,"identity":"053b4f94-281b-4c5a-a9df-0861439e02d3","added_by":"auto","created_at":"2025-09-19 13:50:45","extension":"html","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":83268,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/17d7ecaea841c744f433a228.html"},{"id":91718348,"identity":"4f481de3-9025-4c52-90eb-2286820440c4","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":123463,"visible":true,"origin":"","legend":"\u003cp\u003eAn illustration of the recommendation\u003c/p\u003e\n\u003cp\u003eIn this figure, five people including You rated three items at 5-point scale. Then, four people chose one of the two targets. Majority of people chose T_1, a similar person (i.e., X_3) chose T_2, and a dissimilar person (i.e., X_2) chose T_1. Based on their choice, You would decide which to choose.\u003c/p\u003e","description":"","filename":"Figure1.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/8a6a1a3c2a444da158375e7b.png"},{"id":91720147,"identity":"4f058c7c-e435-4187-85f6-7f0d7af40198","added_by":"auto","created_at":"2025-09-19 14:06:45","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":253457,"visible":true,"origin":"","legend":"\u003cp\u003eAn illustration of one simulation\u003c/p\u003e\n\u003cp\u003eThis figure illustrates the procedure of one simulation. (1) We select 250 group members from all the people (=14,000). (2) We randomly allocate jokes to either training or test trials. (3) We select the targeted person (“You”). (4) We pick up “Dissimilar Person”: the person who has the most dissimilar taste compared with You of all group members. We conduct this procedure using the Pearson correlation in training trials (5, 10, 15...75 trials). (5) You and the Dissimilar Person experienced test trials. The test trials consist of binary choice problems from all jokes in test trials. We test whether the Dissenting preference strategy can predict Your future satisfaction in the test trials. In this illustration, the Dissenting preference strategy can predict Your future satisfaction. We created 300 binary choice problems by combination (= 25 * 24 / 2).\u003c/p\u003e","description":"","filename":"Figure2.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/95c1a80c66d6902660567f80.png"},{"id":91718350,"identity":"c0f4608e-7111-473e-a124-babe8f02c693","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":243599,"visible":true,"origin":"","legend":"\u003cp\u003eResults of main simulations\u003c/p\u003e\n\u003cp\u003e(a) Illustration of main simulation. We can see that the larger the number of trials, the higher the mean accuracy. Especially from 20 to 25 trials, the mean accuracy exceeded the chance level. (b) In addition to the Dissenting preference strategy, this figure displays the results of the Doppelgänger strategy. The Doppelgänger strategy recorded higher accuracy than the Dissenting preference strategy in any number of trials. (c) Results of the relationship between the correlation in the test trials and the number of training trials for the Dissenting preference. The larger the number of trials, the lower the correlation.\u003c/p\u003e","description":"","filename":"Figure32.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/821c15f6137f822b6b18d987.png"},{"id":91718353,"identity":"c4b64bde-1235-40a2-90bc-37a3b84d86bb","added_by":"auto","created_at":"2025-09-19 13:42:45","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":888622,"visible":true,"origin":"","legend":"\u003cp\u003eResults of analysis on individual differences\u003c/p\u003e\n\u003cp\u003eFigure 4a (b) shows the results of the simulations in the Dissenting preference (Doppelgänger) strategy. In (a), the mean accuracy tended to be high in large Dispersion in taste similarity. This tendency was mitigated by high Mean taste similarity. The mean accuracy tended to be especially low in low Mean taste similarity. In Figure 4(c), we subtracted the mean accuracy in the Doppelgänger strategy from the mean accuracy in the Dissenting preference strategy. When Mean taste similarity was low (e.g., -0.1), the Dissenting preference strategy could record higher accuracy than the Doppelgänger strategy: some points were represented in orange to red.\u003c/p\u003e","description":"","filename":"Figure42.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/100a5eefb4476a5510e95932.png"},{"id":91718873,"identity":"23c87db6-091d-4584-aa54-9f3676dff107","added_by":"auto","created_at":"2025-09-19 13:50:45","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":140911,"visible":true,"origin":"","legend":"\u003cp\u003eResults of analysis on group size\u003c/p\u003e\n\u003cp\u003eThis figure indicates the relationship between the mean accuracy (y-axis) and the group size (x-axis). The figure shows that the larger the group sizes, the higher the mean accuracy. Especially from the size of 10 to 20, the mean accuracy became above the chance level.\u003c/p\u003e","description":"","filename":"Figure5.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/cb969193d8bac59414e3b559.png"},{"id":91718872,"identity":"18e4f243-da83-4cde-bb2a-9fc30bbe9c44","added_by":"auto","created_at":"2025-09-19 13:50:45","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":131923,"visible":true,"origin":"","legend":"\u003cp\u003eResults of analysis on dissimilar people.\u003c/p\u003e\n\u003cp\u003e(a) This figure represents the relationship between the mean accuracy and the number of Dissimilar people. The line of the mean accuracy is concave down: The mean accuracy increased up to four with group size, though slightly and then it decreased.\u003c/p\u003e\n\u003cp\u003e(b) This figure analysed the results. Especially, in this figure, we calculated the mean accuracy in all group members and then sorted them according to the rank of dissimilarity. We can see that the relationship between the rank and the mean accuracy was inversely proportional.\u003c/p\u003e","description":"","filename":"Figure6.png","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/074281764374ac5eaaf82613.png"},{"id":92865612,"identity":"8cc8f691-bf50-4e96-a9cf-14d9b7c4d6c2","added_by":"auto","created_at":"2025-10-06 13:02:14","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1880787,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7636367/v1/a0fbb453-0c58-4664-b19c-0597f393a0c2.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"How can we accurately predict for matter of taste using opinions from dissimilar individuals?","fulltext":[{"header":"Introduction","content":"\u003cp\u003eWhich movie should I watch next? Where should I go for today\u0026rsquo;s dinner? We make decisions about matters of taste on a daily basis, on which there are no universal, objective criteria. In these cases, we routinely consider the opinions of others who already experienced and evaluated the options. Then, can we make better decisions to consider them? Studies in social learning, advice-taking etc.,\u003csup\u003e1\u0026ndash;9\u003c/sup\u003e have examined this issue and proposed a formal model of social learning. However, these studies mainly investigated matters of fact which has objective criterion. In other words, they did not consider matters of taste, for which there are no objective criteria.\u003c/p\u003e\u003cp\u003eImportantly, some recent studies\u003csup\u003e\u003cspan additionalcitationids=\"CR11 CR12 CR13 CR14\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e have addressed the matter of taste. Especially a significant study by Analytis et al.(2018)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e proposed several models of social learning strategies based on recommender system algorithms and those about matters of fact and when and how these different strategies thrive or fail. Figure\u0026nbsp;1 illustrates the case where people will rely on others\u0026rsquo; opinions. Imagine that you have already experienced three albums and are thinking about which album to buy or stream. Now, you can use the opinions of four people who have also experienced the albums. Moreover, they show the preferences of the two albums you are considering.\u003c/p\u003e\u003cp\u003eThen, how should we use others\u0026rsquo;opinions? Simply put, Analytis et al.(2018)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e proposed two strategies: ①Similarity: through experiencing some albums, you can identify who has a similar tastes to you. In Fig.\u0026nbsp;1, X_3 has almost the same taste as you. Therefore, you will choose \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\varvec{T}\\_2\\)\u003c/span\u003e\u003c/span\u003e because s/he preferred T_2 to T_1 (hereafter, called \u0026ldquo;Doppelg\u0026auml;nger\u0026rdquo;). ②Majority: You can also rely on the opinions of the majority, as with the wisdom of crowds. In this case, you will choose \u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\:\\varvec{T}\\_1\\)\u003c/span\u003e\u003c/span\u003e since three of the four others preferred T_1 to T_2. Analytis et al.(2018)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e reported that they could work effectively. In other words, these strategies could predict your future satisfaction above the chance level.\u003c/p\u003e\u003cp\u003eHowever, in daily life, we often can not get similar other\u0026rsquo;s opinions and many others\u0026rsquo; opinions. Can we make better decisions even in this case? In this respect, descriptive studies in social and consumer psychology have pointed out the importance of receiving the opinion of \u0026ldquo;dissimilar\u0026rdquo; others. Although some studies show that the opinion of dissimilar others is ignored or discounted, others have indicated that in forming opinions, people take irrelevant cues into account to the extent\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e,\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. In particular, recent studies showed that the opinions of dissimilar people made a person make contrasting choices. For example, when there are items A and B, and the other you dislike, choose item A; you will like the contrast item. In other words, you will choose item B.\u003c/p\u003e\u003cp\u003eIn addition, some studies have shown that contrast choice emerged as a cognitive process (i.e., not by disliking). Tuk et al.(2019)\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e demonstrated that perceptions of dissimilarity trigger a more general hypothesis of dissimilarity. In the experiment (Ex. 1-a), participants read texts about other who had different sex, different taste of jokes, and did different judgmental decision from the participants. Then, they read additional texts, and the other recommended one of the two experimental tasks. Then, the participants were required to answer which tasks they wanted to do. The results showed that the participants chose the task the others did not recommend.\u003c/p\u003e\u003cp\u003eLet\u0026rsquo;s go back to our illustration. In Fig.\u0026nbsp;1, X_2 has a highly dissimilar taste to you. Therefore, based on the aforementioned studies, you seem to choose T_2 against the recommendation of X_2. However, it remains unclear whether this strategy (hereafter called \u0026ldquo;Dissenting preference strategy\u0026rdquo;) enables people to make better decisions since the previous research only empirically observed people\u0026rsquo;s behaviour.\u003c/p\u003e\u003cp\u003eUsing a series of computer simulations on a large-scale dataset, we investigate this issue. As a dataset, we used the Jester dataset\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e, which consists of 14,000 people\u0026rsquo;s 100 joke ratings. Through the simulations, we first identified whether Dissenting preference strategy enables participants to make better decisions. In particular, we point to the significant role of the number of experiences and the size of the neighbourhood. In addition, we compared the Dissenting x preference strategy with the other strategy based on the similarity (Doppelg\u0026auml;nger) and found that the Doppelg\u0026auml;nger strategy usually beat the Dissenting preference strategy. However, we also analysed individual differences to identify those who can leverage the efficacy of the Dissenting preference strategy more than the Doppelg\u0026auml;nger strategy.\u003c/p\u003e\u003cp\u003e-----Figure 1 about here-----\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eAnalysis\u003c/h2\u003e\u003cp\u003eBased on Analytis et al.(2018)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e, we constructed the simulation architecture. Figure\u0026nbsp;2 shows the one trial of the simulations. (1) We select 250 group members from all the people (=\u0026thinsp;14,000). (2) We randomly allocate jokes to either training or test trials. (3) We select the targeted person (\u0026ldquo;You\u0026rdquo;). (4) We pick up the person who has the most dissimilar taste compared with You of all group members (\u0026ldquo;Dissimilar Person\u0026rdquo;). We conduct this procedure using the Pearson correlation in training trials (5, 10, 15...75 trials). (5) You and the Dissimilar Person experienced test trials. The test trials consist of binary choice problems from all jokes in test trials. We created 300 binary choice problems by combination (=\u0026thinsp;25 * 24 / 2). We test whether the Dissenting preference strategy can predict Your future satisfaction in the test trials. Note that the Dissenting preference strategy predicts that You will like the joke that the Dissimilar Person does not like from the binary choice problem. We repeated each step in this procedure, and We ran the simulations 1,000 times for each number of training trials (5, 10, 15...75 trials).\u003c/p\u003e\u003cp\u003eNote that in one trial of the simulations, we tested all 250 people as You. In other words, we repeated (3) ~ (6) for 250 times.\u003c/p\u003e\u003cp\u003e-----Figure 2 about here-----\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eMain results\u003c/h3\u003e\n\u003cp\u003eNotably, in the simulations, we set the number of group members at 250, as in Analytis et al.(2018)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. The assumption is based on the arguments on Dunbar\u0026rsquo;s number\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. The literature proposed that people can build stable relationships up to around 250 because of the cognitive limit. Therefore, the simulations treat the case where You seek opinions around the community, which You can access cognitively.\u003c/p\u003e\u003cp\u003eFigure 3a shows the results of the analysis. This figure indicates the relationship between the mean accuracy and the number of training trials. We can see that the larger the number of trials, the higher the mean accuracy. Especially from 20 to 25 trials, the mean accuracy exceeded the chance level. Therefore, dissimilarity enabled us to make better decisions, especially above 25 trials. Conversely, to make a good decision, dissimilarity (i.e., Dissenting preference strategy) requires people to do training even 25 trials.\u003c/p\u003e\u003cp\u003eFigure 3b adds the result of the additional simulations. In particular, the figure shows the results of the Doppelg\u0026auml;nger strategy, as in Analytis et al.(2018)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. In the additional simulations, the Doppelg\u0026auml;nger strategy chose the person most similar to You in the training trials. In the test trials, the Doppelg\u0026auml;nger strategy predicted that You would like the joke that the most similar person did like from the binary choice problem.\u003c/p\u003e\u003cp\u003eAs with the Dissenting preference strategy, the larger the number of trials, the higher the mean accuracy in the Doppelg\u0026auml;nger strategy. Notably, the Doppelg\u0026auml;nger strategy recorded higher accuracy than the Dissenting preference strategy in any number of trials. Moreover, the Doppelg\u0026auml;nger strategy beat the Dissenting preference strategy when the number of trials in the Doppelg\u0026auml;nger strategy was minimum (=\u0026thinsp;5) and that in the Dissenting preference strategy was maximum (=\u0026thinsp;100). In addition, contrary to the Dissenting preference strategy, the Doppelg\u0026auml;nger strategy recorded above the chance level at any number of trials.\u003c/p\u003e\u003cp\u003eThen, how did the results emerge? Fig.\u0026nbsp;3c indicates the results of further analysis. The analysis examined the relationship between the correlation in the test trials and the number of training trials for the Dissenting preference. The correlation was between You and a Dissimilar Person. First, in Dissenting preference strategy, the larger the number of trials, the lower the correlation. The results indicated that many trials enabled You to pick up a fully dissimilar person. Notably, the correlation became below zero when the number of trials was 25. Thus, we confirmed that the value of the correlation determined the accuracy of the Dissenting preference strategy. As explained in the Analysis section, the Dissenting preference strategy predicts that You will like the joke that a Dissimilar Person does not like.\u003c/p\u003e\u003cp\u003eOverall, we found that people could make good decisions based of dissimilarity, but dissimilarity\u0026rsquo;s effectiveness was relatively low and limited. We shall make a note of this issue in the Discussion section.\u003c/p\u003e\u003cp\u003e-----Figure 3 about here-----\u003c/p\u003e\n\u003ch3\u003eIndividual differences\u003c/h3\u003e\n\u003cp\u003eHereafter, we investigated individual differences in Dissenting preference strategy. In particular, we aimed to identify those who can leverage the Dissenting preference strategy more than the Doppelg\u0026auml;nger strategy.\u003c/p\u003e\u003cp\u003eIn this respect, Analytis et al.(2018)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e proposed two factors that play a role in individual differences: mean taste similarity across all other individuals (hereafter called \u0026ldquo;Mean taste similarity\u0026rdquo;) and variance across all individuals in taste similarity (\u0026ldquo;Dispersion in taste similarity\u0026rdquo;). For example, Analytis et al.(2018)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e demonstrated that the accuracy of the Doppelg\u0026auml;nger strategy was high for those with high Mean taste similarity and high Dispersion in taste similarity. Based on Analytis et al.(2018)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e, we first calculated Mean taste similarity and Dispersion in taste similarity with all other people (=\u0026thinsp;13,999 people) and then conducted the simulations. We conducted the simulations by regarding each individual as You. In this simulation, we set the number of trials at a maximum (=\u0026thinsp;75). Except for these assumptions, the simulations were the same as the main simulations.\u003c/p\u003e\u003cp\u003eFigure 4a shows the results of the simulations in Dissenting preference strategy. The mean accuracy tended to be high in large Dispersion in taste similarity. When the Dispersion in taste similarity is large, You can pick up the person with a dissimilar taste. This tendency was mitigated by high Mean taste similarity. The mean accuracy tended to be especially low in low Mean taste similarity. This is because when Mean taste similarity is high, it is relatively hard for You to find the dissimilar person.\u003c/p\u003e\u003cp\u003eFigure 4b shows the results of the simulations in the Doppelg\u0026auml;nger strategy. The Doppelg\u0026auml;nger strategy seeks a person who has a similar taste to You. Therefore, contrary to the Dissenting preference strategy, Mean taste similarity should be high to achieve high accuracy. Next, similar to the Dissenting preference strategy, Dispersion in taste similarity should also be high. This is because, in this case, there seems to be a person with an exceptionally similar taste to You.\u003c/p\u003e\u003cp\u003eIn Fig.\u0026nbsp;4c, we compared the Dissenting preference strategy with the Doppelg\u0026auml;nger strategy. Especially for each individual, we subtracted the mean accuracy in the Doppelg\u0026auml;nger strategy from the mean accuracy in the Dissenting preference strategy. The value was above zero for individuals who recorded higher accuracy in the Dissenting preference strategy than in the Doppelg\u0026auml;nger strategy. Although the results were ambiguous for Dispersion in taste similarity, we can see a clear tendency for Mean taste similarity. When Mean taste similarity was low (e.g., -0.1), some points were represented in orange to red. For the individuals, the Dissenting preference strategy recorded higher accuracy than the Doppelg\u0026auml;nger strategy. The results are derived from the following findings. As mentioned above, in the Dissenting preference strategy, the mean accuracy tended to be high with low Mean taste similarity. On the contrary, in the Doppelg\u0026auml;nger strategy, the mean accuracy tended to be low in low Mean taste similarity. Therefore, in low Mean taste similarity, the Dissenting preference strategy could beat the Doppelg\u0026auml;nger strategy for some individuals.\u003c/p\u003e\u003cp\u003eIn summary, we found conditions under which the Dissenting preference strategy could be highly accurate. Moreover, through the comparison, we identified situations under which the Dissenting preference strategy could record higher accuracy than the Doppelg\u0026auml;nger strategy (i.e., Mean Taste Similarity was low).\u003c/p\u003e\u003cp\u003e-----Figure 4 about here-----\u003c/p\u003e\n\u003ch3\u003eAdditional analysis 1: Group size\u003c/h3\u003e\n\u003cp\u003eIn the main simulations, we set the group size where an individual could seek opinions at 250. This assumption was based on Dunbar\u0026rsquo;s number\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u003c/sup\u003e. However, we do not get the exact number of people an individual can rely on. Therefore, it is worth manipulating the group size and examining the effectiveness of the Dissenting preference strategy. Here, we set group sizes at 4, 10, 20, 50, 100, 250, 500, 1000. For simplicity, we set the number of trials at 75. Except for the assumptions, we run the simulations as with the main simulations.\u003c/p\u003e\u003cp\u003eFigure 5 shows the results of the simulations. This figure indicates the relationship between the mean accuracy and the group size. We can see that the larger the group sizes, the higher the mean accuracy. Especially from the size of 10 to 20, the mean accuracy became above the chance level. Therefore, dissimilarity enabled us to make good decisions, especially above 20. In anthropology, a band society\u003csup\u003e\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e was proposed as the simplest form of human society. The band generally consists of an extended family or clan. The size of the band is estimated at 30. Therefore, the results implied that dissimilarity enabled us to make good decisions in the band society.\u003c/p\u003e\u003cp\u003eThis figure also shows the result of the Doppelg\u0026auml;nger strategy. As the figure shows, the larger the group sizes, the higher the mean accuracy. In addition, the Doppelg\u0026auml;nger strategy was recorded above the chance level at any number of trials, contrary to the Dissenting preference strategy. Moreover, the Doppelg\u0026auml;nger strategy recorded higher accuracy than the Dissenting preference strategy in any group size. Importantly, the Doppelg\u0026auml;nger strategy beat the Dissenting preference strategy when the group size in the Doppelg\u0026auml;nger strategy was minimum (=\u0026thinsp;4) and that in the Dissenting preference strategy was maximum (=\u0026thinsp;1,000). Thus, we found the effectiveness and the limitation of dissimilarity for the group size.\u003c/p\u003e\u003cp\u003e-----Figure 5 about here-----\u003c/p\u003e\n\u003ch3\u003eAdditional analysis 2: Dissimilar people\u003c/h3\u003e\n\u003cp\u003eHow can we enhance the effectiveness of the Dissenting preference strategy? Analytis et al.(2018)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e indicated that increasing the number of people you rely on can enhance the prediction strategy. Especially by taking 10 of the most similar people, Analytis et al.(2018)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e showed that prediction accuracy improved by more than 5%. Based on the findings, we conducted additional simulations that took Dissimilar people beyond one person. In particular, we set the number of most Dissimilar people at 1 (i.e., same as main simulations), 2, 3, 4, 6, 10, 20, 50, and 100.\u003c/p\u003e\u003cp\u003eNote that preference prediction was conducted using the majority rule\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e,\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. For example, when Dissimilar people were 6, and four people liked A and two people liked B, the judge by the Dissimilar people was A. therefore, the preference prediction was correct when the target liked B more than A. If a half of Dissimilar people liked A and another half of Dissimilar people liked B, the strategy predicted the preference randomly.\u003c/p\u003e\u003cp\u003eFigure 6a shows the results of the additional simulations. This figure indicates the relationship between the mean accuracy and the number of Dissimilar people. As the figure displays, the line of the mean accuracy is concave down: The mean accuracy increased up to three with group size, though slightly and then it decreased. In other words, we could identify an optimal number of Dissimilar people.\u003c/p\u003e\u003cp\u003eThis figure also shows the results of the Doppelg\u0026auml;nger strategy. We can also see that the Doppelg\u0026auml;nger strategy's mean accuracies were inverted-U shape. The mean accuracy increased to fifty with group size, though slightly, then decreased.\u003c/p\u003e\u003cp\u003eThen, how did the results emerge? We especially focused on the findings that the mean accuracies were inverted-U shape for Dissimilar people in the Dissenting preference strategy. Figure\u0026nbsp;6b displays the results of the analysis. This analysis calculated the mean accuracy in all group members and then sorted them according to the rank of dissimilarity. We can see that the relationship between the rank and the mean accuracy was inversely proportional. Particularly in Rank 6, the accuracy was below the chance level. Therefore, adding a small number of members could enhance the accuracy; however, adding a large number of people decreased the accuracy.\u003c/p\u003e\u003cp\u003e-----Figure 6 about here-----\u003c/p\u003e"},{"header":"Discussion","content":"\u003cp\u003eIn everyday decisions, we routinely consider the opinions of others who already experienced them. It is known that relying on similar others or many others enables us to make better decisions. However, we often cannot rely on them. In this case, how can we leverage other\u0026rsquo;s opinions? Based on the behavioural studies, we focused on the strategy of dissimilarity: One should choose a different opinion from a dissimilar person chose. Especially in the case where there are binary options A and B, and a dissimilar person chose A, one should choose B.\u003c/p\u003e\u003cp\u003eTo test the strategy's performance, we conducted computer simulations. In the simulations, we picked an individual who was ready to choose binary options and another person who had a dissimilar taste to the individual. Then, we tested the strategy's performance. The results showed that the strategy of dissimilarity could accurately predict the individual\u0026rsquo;s future satisfaction.\u003c/p\u003e\u003cp\u003eThen, we compared the strategy in the previous study: the strategy of similarity. As a result, the strategy of dissimilarity was largely beaten by this strategy. Therefore, we attempted to identify the conditions where the strategy of dissimilarity performed better. As a result, we found that for those who had different tastes from others, the strategy of dissimilarity could be performed better than the strategy of similarity. Moreover, we examined the situation where the strategy of dissimilarity utilised multiple people. As a result, we identified the optimal number for the people (=\u0026thinsp;4).\u003c/p\u003e\u003cp\u003eThrough the simulations, we also clarified the conditions under which the strategy of dissimilarity performed well: a moderate number of training trials and a moderate number of members were needed for the strategy to accurately predict one\u0026rsquo;s future satisfaction.\u003c/p\u003e\u003cp\u003eWhile many previous studies on social learning treated as a matter of fact\u003csup\u003e\u003cspan additionalcitationids=\"CR2 CR3 CR4 CR5 CR6 CR7 CR8\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, this paper focused on the matter of taste\u003csup\u003e\u003cspan additionalcitationids=\"CR11 CR12\" citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e, for which there is no objective criteria, along with Analytis et al.(2018)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. Importantly, to the best of our knowledge, this is the first study that showed \u0026ldquo;dissimilarity\u0026rdquo; could lead to better decisions. In other words, our study implied that dissimilarity could become \u0026ldquo;information\u0026rdquo; for those who are ready to make decisions.\u003c/p\u003e\u003cp\u003eAs we mentioned in the Introduction, studies in social and consumer psychology have pointed out the importance of receiving the opinion of \u0026ldquo;dissimilar\u0026rdquo; others. Especially, Tuk et al.(2019)\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e showed that we chose different option form the dissimilar person. Combined with the findings in our study, the strategy may be \u0026ldquo;adaptive\u0026rdquo;\u003csup\u003e22,23\u003c/sup\u003e. However, it should be noted that our findings\u0026mdash;\u0026mdash;the strategy of dissimilarity could predict the individual\u0026rsquo;s future satisfaction\u0026mdash;\u0026mdash;have several limitations: a moderate number of training trials and a moderate number of members were needed. Therefore, future studies which examine whether the strategy of dissimilarity performs well directly should be needed.\u003c/p\u003e\u003cp\u003eAnother promising approach is to test the findings in our study could extend a matter of fact. For example, when an individual sees that another person keeps making mistakes in social learning\u003csup\u003e\u003cspan additionalcitationids=\"CR2 CR3 CR4 CR5 CR6 CR7 CR8\" citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e, the individual might make accurate decision by making different decision from another person. It is important to examine this point, including how people actually engage in social learning.\u003c/p\u003e\u003cp\u003eMoreover, our findings could contribute to real-world review sites. Online review sites have usually utilised recommender systems\u003csup\u003e\u003cspan additionalcitationids=\"CR24 CR25 CR26 CR27 CR28\" citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e\u003c/sup\u003e (e.g., content-based, demographic-based and personality-based recommender systems). They mainly exploit the users who have similar pattern to the target user. However, this study demonstrated that participants could leverage other users who have dissimilar preference patterns. Therefore, we may build new recommender system based of dissimilarity.\u003c/p\u003e\u003cp\u003eImportantly, the recommender system based of dissimilarity could complement usual recommender systems. Analytis et al.(2023)\u003csup\u003e\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e\u003c/sup\u003e tested representative collaborative filtering algorithms and found the \u0026ldquo;Mainstream-taste bias\u0026rdquo;: they perform much worse for non-mainstream or unusual tastes than mainstream users. In this respect, as we mentioned in the Individual Differences section in Results, the strategy of dissimilarity performed well for those with unusual tastes. Therefore, one promising approach is to construct \u0026ldquo;hybrid\u0026rdquo; recommender systems. Put simply, by combining collaborative filtering with it, we might enhance the quality of the recommender system for various types of people.\u003c/p\u003e\u003cp\u003eOne limitation of this study is the architecture of the simulation. In the test trials, a target and another person (people) made only a binary choice. On the contrary, in the real world, we choose an option from a wide range of options. Therefore, it is important to test the strategy for dissimilarity in a wide range of choice situations. In the future, we will re-conduct simulations when there are more options.\u003c/p\u003e\u003cp\u003eAnother limitation was as for stimuli: This study focused on joke, as with the previous study\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e. However, it remains unclear whether the findings of this study could apply to the everyday\u0026rsquo;s consumer behaviour, as illustrated in Introduction. Therefore, future studies which examine another range of data should be conducted\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e\u003ch2\u003eDataset\u003c/h2\u003e\u003cp\u003eAs with Analytis et al.(2018)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e, we utilised The Jester dataset for a large-scale dataset. In this site (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://eigentaste.berkeley.edu\u003c/span\u003e\u003cspan address=\"http://eigentaste.berkeley.edu\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.), people voluntarily read and rate jokes on a scale ranging from not funny (\u0026minus;\u0026thinsp;10) to funny (+\u0026thinsp;10). In other words, people rated the jokes on a scale of 20. Along with Analytis et al.(2018)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e, again, we only used the data of participants who evaluated all jokes (reducing the number of participants from 73,421 to 14,116). Specifically, we analysed data from which Analytis et al.(2018)\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e randomly selected 14,000 people from the 14,116 people.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eAnalysis of (dis) similarity\u003c/h2\u003e\u003cp\u003eWe used the Pearson correlation coefficient as a measure of similarity between two individuals or two items where (dis)similarity between two individuals \u003cem\u003ei\u003c/em\u003e and \u003cem\u003ej\u003c/em\u003e is defined as\u003cdiv id=\"Equa\" class=\"Equation\"\u003e\u003cdiv format=\"TEX\" class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e\n$$\\:w(i,\\:j)\\:=\\:\\frac{{\\sum\\:}_{n\\:=\\:1}^{M}({u}_{in}\\:-\\:{ū}_{i})({u}_{jn}\\:-\\:{\\text{ū}}_{\\text{j}})}{\\sqrt{{\\sum\\:}_{n\\:=\\:1}^{M}{({u}_{in}\\:-\\:{\\text{ū}}_{\\text{i}})}^{2}}\\sqrt{{\\sum\\:}_{n\\:=\\:1}^{M}{({u}_{jn}\\:-\\:{\\text{ū}}_{\\text{j}})}^{2}}}$$\u003c/div\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eFor the similar options strategy, (dis) similarity is calculated using the Pearson correlation coefficient but between two items \u003cem\u003ek\u003c/em\u003e and \u003cem\u003el\u003c/em\u003e. \u003cem\u003eM\u003c/em\u003e stands for the total number of options or people and n for specific options or people. We chose to adopt the Pearson correlation because in the recommender system literature, it is the canonical measure.\u003c/p\u003e\u003c/div\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003ch2\u003eDECLARATION OF INTERESTS\u003c/h2\u003e\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFUNDING\u003c/h2\u003e\u003cp\u003eThis research received no external funding. We used only open data and simulation analysis.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eIF and YK developed the study concept and contributed to the study design. IF conduct the simulations and wrote the manuscript, with feedback from KY.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eWe thank Kazuhiro Ueda for their helpful comments.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe R-code and the datasets analysed during the current study are available in the Mendeley Data: https://data.mendeley.com/datasets/6s46dfytvp/1\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eBoyd, R. \u0026amp; Richerson, P. J. \u003cem\u003eCulture and the Evolutionary Process\u003c/em\u003e (Univ. Chicago, 1985).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eClemen, R. T. Combining forecasts: a review and annotated bibliography. \u003cem\u003eInt. J. Forecast.\u003c/em\u003e \u003cb\u003e5\u003c/b\u003e, 559\u0026ndash;583 (1989).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKameda, T. \u0026amp; Nakanishi, D. Cost\u0026ndash;benefit analysis of social/cultural learning in a nonstationary uncertain environment: an evolutionary simulation and an experiment with human subjects. \u003cem\u003eEvol. Hum. Behav.\u003c/em\u003e \u003cb\u003e23\u003c/b\u003e, 373\u0026ndash;393 (2002).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKameda, T. \u0026amp; Nakanishi, D. Does social/cultural learning increase human adaptability? Rogers\u0026rsquo;s question revisited. \u003cem\u003eEvol. Hum. Behav.\u003c/em\u003e \u003cb\u003e24\u003c/b\u003e, 242\u0026ndash;260 (2003).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLaland, K. N. Social learning strategies. \u003cem\u003eAnim. Learn. Behav.\u003c/em\u003e \u003cb\u003e32\u003c/b\u003e, 4\u0026ndash;14 (2004).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eHastie, R. \u0026amp; Kameda, T. The robust beauty of majority rules in group decisions. \u003cem\u003ePsychol. Rev.\u003c/em\u003e \u003cb\u003e112\u003c/b\u003e, 494\u0026ndash;508 (2005).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eLarrick, R. P., Mannes, A. E. \u0026amp; Soll, J. B. in Frontiers in Social Psychology:Social Judgment and Decision Making (ed. Krueger, J. I.) 227\u0026ndash;242Psychology Press, New York, NY, (2012).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGoldstone, R. L., Wisdom, T. N., Roberts, M. E. \u0026amp; Frey, S. Learning along with others. \u003cem\u003ePsychol. Learn. Motiv\u003c/em\u003e. \u003cb\u003e58\u003c/b\u003e, 1\u0026ndash;45 (2013).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWitt, A., Toyokawa, W., Lala, K. N., Gaissmaier, W. \u0026amp; Wu, C. M. Humans flexibly integrate social information despite interindividual differences in reward. \u003cem\u003eP Natl. Acad. Sci.\u003c/em\u003e \u003cb\u003e121\u003c/b\u003e, e2404928121 (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAnalytis, P. P., Barkoczi, D. \u0026amp; Herzog, S. M. Social learning strategies for matters of taste. \u003cem\u003eNat. Hum. Behav.\u003c/em\u003e \u003cb\u003e2\u003c/b\u003e, 415\u0026ndash;424 (2018).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eYaniv, I., Choshen-Hillel, S. \u0026amp; Milyavsky, M. Receiving advice on matters of taste: Similarity, majority influence, and taste discrimination. \u003cem\u003eOrgan. Behav. Hum. Decis. Process.\u003c/em\u003e \u003cb\u003e115\u003c/b\u003e, 111\u0026ndash;120 (2011).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eM\u0026uuml;ller-Trede, J., Choshen-Hillel, S., Barneron, M. \u0026amp; Yaniv, I. The wisdom of crowds in matters of Taste. \u003cem\u003eManage. Sci.\u003c/em\u003e \u003cb\u003e64\u003c/b\u003e, 1779\u0026ndash;1803 (2017).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBarneron, M., Allalouf, A. \u0026amp; Yaniv, I. Rate it again: Using the wisdom of many to improve performance evaluations. \u003cem\u003eJ. Behav. Decis. Mak.\u003c/em\u003e \u003cb\u003e32\u003c/b\u003e, 485\u0026ndash;492 (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFujisaki, I., Honda, H. \u0026amp; Ueda, K. A simple cognitive method to improve the prediction of matters of taste by exploiting the within-person wisdom-of-crowd effect. \u003cem\u003eSci. Rep.\u003c/em\u003e \u003cb\u003e12\u003c/b\u003e, 12413 (2022).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eKopsacheilis, O. et al. Crowdsourcing the assessment of wine quality: Vivino ratings, professional critics, and the weather. \u003cem\u003eJ. Wine Econ.\u003c/em\u003e \u003cb\u003e19\u003c/b\u003e, 285\u0026ndash;304 (2024).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eSchwarz, N. \u0026amp; Clore, G. L. Mood, misattribution, and judgments of well-being: Informative and directive functions of affective states. \u003cem\u003eJ. Pers. Soc. Psychol.\u003c/em\u003e \u003cb\u003e45\u003c/b\u003e, 513\u0026ndash;523 (1983).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eMeyers-Levy, J. \u0026amp; Sternthal, B. A 2-factor explanation of assimilation and contrast effects. \u003cem\u003eJ. Market Res.\u003c/em\u003e \u003cb\u003e30\u003c/b\u003e, 359\u0026ndash;368 (1993).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eTuk, M. A., Verlegh, P. W., Smidts, A. \u0026amp; Wigboldus, D. H. You and I have nothing in common: The role of dissimilarity in interpersonal influence. \u003cem\u003eOrgan. Behav. Hum. Decis. Process.\u003c/em\u003e \u003cb\u003e151\u003c/b\u003e, 49\u0026ndash;60 (2019).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eDunbar, R. \u003cem\u003eHow Many Friends Does One Person Need? Dunbar\u0026rsquo;s Number and Other Evolutionary Quirks\u003c/em\u003e (Faber \u0026amp; Faber, 2010).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eWilliams, B. J. \u0026amp; Wobst, H. M. \u003cem\u003eA model of band society\u003c/em\u003e29i\u0026ndash;138 (Memoirs of the Society for American Archaeology, 1974).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGigerenzer, G., Todd, P. M., Group, A. B. C. R. \u0026amp; T \u003cem\u003eSimple heuristics that make us smart\u003c/em\u003e (Oxford University Press, 2000).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eGoldstein, D. G. \u0026amp; Gigerenzer, G. Models of ecological rationality: the recognition heuristic. \u003cem\u003ePsychol. Rev.\u003c/em\u003e \u003cb\u003e109\u003c/b\u003e (1), 75\u0026ndash;90 (2002).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003ePazzani, M. \u0026amp; Billsus, D. in \u003cem\u003ein The Adaptive Web\u003c/em\u003e. 325\u0026ndash;341 (eds Brusilovsky, P., Kobsa, A. \u0026amp; Nejdl, W.) (Springer, 2007).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eNunes, M. A. S. \u0026amp; Hu, R. Personality-based recommender systems: an overview. In Proc. 6th ACM Conference on Recommender Systems 5\u0026ndash;6ACM, (2012).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eBobadilla, J., Ortega, F., Hernando, A. \u0026amp; Guti\u0026eacute;rrez, A. Recommender systems survey. \u003cem\u003eKnowl. Based Syst.\u003c/em\u003e \u003cb\u003e46\u003c/b\u003e, 109\u0026ndash;132 (2013).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eFern\u0026aacute;ndez-Tobas, I., Braunhofer, M., Elahi, M., Ricci, F. \u0026amp; Cantador, I. Alleviating the new user problem in collaborative filtering by exploiting personality information. \u003cem\u003eUser Model. User Adapt. Interact.\u003c/em\u003e \u003cb\u003e26\u003c/b\u003e, 221\u0026ndash;255 (2016).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAnalytis, P. P., Barkoczi, D., Lorenz-Spreen, P. \u0026amp; Herzog, S. The structure of social influence in recommender networks. In: Proceedings of the web conference 2020, 2655\u0026ndash;61. WWW \u0026rsquo;20. New York, NY, USA: association for computing machinery (2020).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAnalytis, P. P. \u0026amp; Hager, P. Collaborative filtering algorithms are prone to mainstream-taste bias. In Proceedings of the 17th ACM Conference on Recommender Systems (pp. 750\u0026ndash;756). (2023).\u003c/span\u003e\u003c/li\u003e\u003cli\u003e\u003cspan\u003eAnalytis, P. P., Kaushik, K., Herzog, S., Bahrami, B. \u0026amp; Deroy, O. A recommender network perspective on the informational value of critics and crowds. (2024). arXiv preprint arXiv:2403.18868.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Social learning, Online recommendation, Matter of taste, Dissimilarity","lastPublishedDoi":"10.21203/rs.3.rs-7636367/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7636367/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003ePeople frequently rely on the opinions of others to make decisions regarding matters of taste, such as choosing movies or restaurants. Then, how can we get accurate prediction from others' opinions? In this respect, similarity and majority-based strategies have been studied. However, the potential benefits of relying on dissimilar individuals' opinions remain unclear. This study investigates the efficacy of a novel strategy, termed the \"Dissenting preference strategy\". This strategy involves making a different (or opposite) choice from that of a person whose preferences differ from one's own. We used computer simulations based on the dataset comprising ratings from 14,000 individuals. Our results reveal that this strategy can improve decision-making accuracy, particularly when a moderate number of experiences (25 or more) and a certain neighborhood size (around 20 individuals or more) are available. However, the similarity-based strategy (\"Doppelg\u0026auml;nger strategy\") consistently outperformed the Dissenting preference strategy. Nevertheless, further analyses highlighted that individuals with lower mean taste similarity could benefit more from the Dissenting preference strategy than from Doppelg\u0026auml;nger strategy. Additionally, incorporating multiple dissimilar individuals (up to four) slightly enhanced the accuracy of the strategy. These findings underscore the conditional utility of dissimilar people in decision-making and suggest avenues for integrating dissimilarity into recommendation systems, especially for users with atypical preferences.\u003c/p\u003e","manuscriptTitle":"How can we accurately predict for matter of taste using opinions from dissimilar individuals?","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-09-19 13:42:40","doi":"10.21203/rs.3.rs-7636367/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"b09b7b89-5806-4a12-b1e1-1fa1fe966f9d","owner":[],"postedDate":"September 19th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":54921133,"name":"Physical sciences/Mathematics and computing"},{"id":54921134,"name":"Biological sciences/Psychology"},{"id":54921135,"name":"Social science/Psychology"}],"tags":[],"updatedAt":"2025-10-06T12:54:04+00:00","versionOfRecord":[],"versionCreatedAt":"2025-09-19 13:42:40","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7636367","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7636367","identity":"rs-7636367","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}