Session-based Recommendation Systems with Graph ATtention Networks

Abstract

Recommender systems give users beneficial product or service recommendations for their decision-making processes. Today, a variety of application domains such as YouTube, Amazon, Facebook, and ResearchGate have proven the validity of classic recommendation systems that employ collaborative and content-based filtering techniques. Session-based recommender systems, a novel RS paradigm, have evolved in recent years in order to give timelier and more accurate next-item recommendations that are responsive to being adjusted in various session circumstances. SBRSS strives to record dynamic and short-term user preferences within sessions. The literature only includes a few models with poor precision and efficacy as proposed development methodologies for SBRS models, which are this type of system's primary objectives. The goal of this thesis is to explicitly provide a new deep learning design for session-based recommender systems based on graph neural networks (GNNs) via the intriguing architecture of graph attention networks (GAT). Currently, gat-based methodologies are among the most cutting-edge techniques used in many research fields, and SBRs can take advantage of them to greatly research fields, and SBRs can take advantage of them to greatly enhance the outcomes of their suggestions