Abstract
The article proposes a model for data in recommendation in recommender systems, which is based on the implementation of fuzzy logic in the collaborative filtering method to improve the quality of personalized recommendations. Particular attention is paid to the problems of data sparsity, uncertainty of user ratings, and the subjectivitye of interpretation of criteria, which traditionally complicates the work of classical recommendation algorithms. The study substantiates the feasibility of using personalized triangular membership functions, which allow for the reflectingon of the personal preferences and evaluation characteristics of each user. A formalized procedure for constructing and dynamically updating the parameters of such functions for all evaluation criteria is proposed. The Mamdani method was used to calculate the degree of similarity between users, which takes into account the fuzziness of ratings and allows logical conclusions to be drawn based on a system of rules. This approach makes it possible to determine the level (degree) of similarity between users, taking into account multidimensional criteria and their qualitative interpretation. In addition, the procedure for defuzzifying the obtained fuzzy similarity values and integrating them into the rating prediction process was demonstrated. To evaluate the effectiveness of the developed model, a model experiment was conducted on an artificially generated dataset with a controlled structure and a given level of sparsity. Metrics based on mean square error (MSE), root mean square error (RMSE), and sum of squares error (SSE) were used to compare the proposed approach with the results of basic collaborative filtering. The results demonstrate the potential of the modified model to reduce prediction error in conditions of incomplete and fuzzy data, as well as to improve the adaptability of recommendations by taking into account individual evaluation models. The proposed approach can be used as a basis for building more robust, flexible, and interpretable next-generation recommendation systems.