Natural Language Processing Techniques to Identify Zoonosis Awareness

In this study, we incorporated several NLP techniques to identify the most important factors in the open-ended responses part of the Knowledge, Attitudes, and Practices: Survey of Zoonoses in Wildlife Trade (KAP) in Cambodia. These included: TF-IDF, ngrams, Latent Semantic allocation (LSA), k-means, Latent Dirichlet Allocation (LDA), and Top2Vec. The top topics participants identified included 1) stating that they handled wildlife by setting traps and mist nets, 2) stating they were bitten by bat or rat, 3) which zoonotic symptoms caused sickness, 4) describing how they would go to the hospital when they came down with zoonotic symptoms, and 5) saying that they were aware of avian flu and its symptoms. Based on our findings, recommendations for Cambodian public health officials include: 1) they need to educate participants to wear protective gear to prevent from being bitten by bats and rats during their jobs with these animals, and 2) they need to educate participants about the danger of different types of zoonotic diseases including Ebolavirus, Mojianvirus, etc., so that these participants can recognize the risks when handling bats and rats, and so they can take early action by seeking medical help as soon as they are bitten.