Рекомендація книг з урахуванням емоцій, що містяться в текстових відгуках читачів

The purpose of the article is to study methods for obtaining emotional analysis data from readers’ text reviews for use in the recommendation functions of e-commerce systems that provide online book reading or audiobook listening services. The research methodology consists of systems approach, systems analysis and synthesis methods, and structural modelling of relational databases. The scientific novelty lies in developing a modified method for determining user-based collaborative filtering recommendations, which considers data from the emotional analysis of text reviews of books read. Conclusions. This article considers three approaches to evaluating the emotions of readers’ text reviews. The first approach uses lexicon-based dictionaries, in which sets of emotions index words. Text emotion scores are calculated based on the statistics of the occurrence of words in sets of emotions in the lexicon. The second approach uses the BERT language model, allowing you to obtain emotional scores upon request. The BERT language model is a neural network that provides the ability to take into account the semantic meaning of words in a sentence, including the words to the left and right of them. The third approach uses the ChatGPT transformer bot. The evaluation of emotions in the reviews text can be done in real time using the ChatGPT API function. The output parameters of the API function are the installation text for the ChatGPT sentiment analyser and the review text, and the input parameter is emotion score data in JSON format. Taking into account the considered approaches, a modified user-based collaborative filtering recommendation method has been developed, which takes emotional analysis data into account. For this purpose, it is proposed that the average cosine similarity of rating assessments and the emotional evaluations of users’ textual reviews be considered during the data selection stage. The proposed method of determining recommendations can be used in library systems with an implemented user function of providing reviews of read books.