The improved model of user similarity coefficients computation For recommendation systems

TL;DR

This paper presents an improved model for calculating user similarity coefficients in recommendation systems, which optimizes recommendation list formation time while maintaining or slightly improving accuracy.

Contribution

The paper introduces a novel user similarity coefficient model that accounts for recalculation periods, enhancing efficiency and recommendation quality.

Findings

Increases recommendation system efficiency up to 2 times.

Maintains or slightly improves Precision and Recall.

Model adapts to user preference changes over time.

Abstract

The subject matter of the article is a model of calculating the user similarity coefficients of the recommendation systems. The goal is the development of the improved model of user similarity coefficients calculation for recommendation systems to optimize the time of forming recommendation lists. The tasks to be solved are: to investigate the probability of changing user preferences of a recommendation system by comparing their similarity coefficients in time, to investigate which distribution function describes the changes of similarity coefficients of users in time. The methods used are: graph theory, probability theory, radioactivity theory, algorithm theory. Conclusions. In the course of the researches, the model of user similarity coefficients calculating for the recommendation systems has been improved. The model differs from the known ones in that it takes into account the recalculation period of similarity coefficients for the individual user and average recalculation period of similarity coefficients for all users of the system or a specific group of users. The software has been developed, in which a series of experiments was conducted to test the effectiveness of the developed method. The conducted experiments showed that the developed method in general increases the quality of the recommendation system without significant fluctuations of Precision and Recall of the system. Precision and Recall can decrease slightly or increase, depending on the characteristics of the incoming data set. The use of the proposed solutions will increase the application period of the previously calculated similarity coefficients of users for the prediction of preferences without their recalculation and, accordingly, it will shorten the time of formation and issuance of recommendation lists up to 2 times.