Automatic Suggestion Detection from Online User Reviews with Information Retrieval and Deep Learning

After the outbreak of COVID-19, the use of the web platform as a means of communication between users and decision-making in online shopping has been accepted in e-commerce. Comments of users express substantially positive and negative emotions about a particular item, which offers suggestions for the improvement of products and organizations. Therefore, mining opinions to extract suggestions can increase the utility of companies and organizations. A good deal of research on suggestion mining employs rule-based approaches and statistical classifiers using manually identified features, while recently, several researchers have considered solutions based on deep learning tools and techniques. The purpose of this study is to use information retrieval techniques for the automatic detection of suggestions. Therefore, distance measurement approaches, multilayer perceptron neural networks, support vector machines, regression logistics, convolutional neural networks with TF-IDF, BOW, and Word2Vec vectors, and keyword extraction were adopted. The proposed approach is presented on the SemEval2019 dataset for extracting suggestions from the text of online user reviews. The results show that the F 1 Score has improved, as compared to the previous work, and has reached 0.87. The experiments indicate that information retrieval-based approaches tend to be promising for this task.