MODIFICATION OF THE ITEM-BASED COLLABORATIVE FILTERING METHOD

Abstract

This article presents an improved approach to generating recommendations based on item-based collaborative filtering, which integrates semantic and temporal factors with cluster analysis methods. The main goal of the proposed improvements is to reduce the computational load on the recommendation module and increase the relevance of the generated recommendations by excluding content that has no semantic meaning, while preserving the context when generating recommendations for users. The algorithm uses collaborative filtering methods in combination with the HDBSCAN clustering algorithm, which ensures efficient processing of large amounts of data. The use of semantic and temporal characteristics of objects made it possible to significantly improve the accuracy of user rating approximations and, accordingly, the quality of personalized recommendations. In addition, a method for optimizing data processing was proposed, taking into account the dynamics of changes in user interests over time. The developed system also provides for the possibility of classifying content according to relevant criteria, which contributes to increased filtering accuracy. The preprocessing procedure involves data aggregation followed by clustering, which reduces the computational complexity of generating recommendations. The similarity between objects is calculated taking into account both temporal and semantic characteristics. Software has been developed to test the proposed algorithm on different data sets. Experimental verification of the proposed approach on different datasets has demonstrated its advantages over basic methods in terms of accuracy and performance.