BiLSTCAN: A Novel SRS-Based Bidirectional Long Short-Term Capsule Attention Network for Dynamic User Preference and Next-Item Recommendation

Abstract

Numerous research efforts are endeavoring to boost the performance of dynamic user preferences and next-item recommendations, which are pivotal tasks within sequential recommender systems. It is a challenging research problem in recommender systems. However, the majority of research faces notable hurdles, especially the limitation of unidirectional methods, which can solely be executed in the forward direction. In this paper, we propose a novel sequential recommender system-based bidirectional long short-term capsule attention network (BiLSTCAN). It amalgamates novel bidirectional capsule attention network and bidirectional long short-term memory network. BiLSTCAN integrates attention network and incorporates time updates into bidirectional capsule attention network for capturing dynamic user preference of sequential recommendation tasks. A bidirectional architecture is a promising framework that can concurrently capture information from both the forward and backward directions, yielding optimal performance. Additionally, we utilize bidirectional long short-term memory for predicting next-item recommendation. Empirical experiments on five benchmark datasets illustrate that BiLSTCAN surpasses the compared state-of-the-art methods in every Top-N and every density.