전이 확률 기반 벡터를 이용한 추천 시스템 성능 향상

Abstract

Numerous companies are now able to store and manage huge amounts of information about their customers. Accordingly, studies on recommender systems are actively being conducted to use the information more efficiently. Among them, studies that wish to have high predictability using additional information other than purchase information are presented in this paper with a simple method to reduce costs and increase accuracy. The corresponding module is a vector based on the probability that an item is transferred to another item. Experiments conducted on public datasets show that the performances of the proposed architecture have improved by an average of 9.7% compared to the benchmark models. It was also intended to provide direction for cold-start problem resolution at no additional cost.