Intent-aware Recommendation via Disentangled Graph Contrastive Learning

TL;DR

This paper presents the Intent-aware Recommendation via Disentangled Graph Contrastive Learning (IDCL), which simultaneously learns interpretable intents and behavior distributions over those intents.

Abstract

Graph neural network (GNN) based recommender systems have become one of the mainstream trends due to the powerful learning ability from user behavior data. Understanding the user intents from behavior data is the key to recommender systems, which poses two basic requirements for GNN-based recommender systems. One is how to learn complex and diverse intents especially when the user behavior is usually inadequate in reality. The other is different behaviors have different intent distributions, so how to establish their relations for a more explainable recommender system. In this paper, we present the Intent-aware Recommendation via Disentangled Graph Contrastive Learning (IDCL), which simultaneously learns interpretable intents and behavior distributions over those intents. Specifically, we first model the user behavior data as a user-item-concept graph, and design a GNN based behavior disentangling module to learn the different intents. Then we propose the intent-wise contrastive learning to enhance the intent disentangling and meanwhile infer the behavior distributions. Finally, the coding rate reduction regularization is introduced to make the behaviors of different intents orthogonal. Extensive experiments demonstrate the effectiveness of IDCL in terms of substantial improvement and the interpretability.

Figures (6)

• Figure 1: An illustration of behavior distributions. The intents for each behavior are shown on the arrow (the main intents are bolded).
• Figure 2: The framework of the proposed Intent-aware Recommendation via Disentangled Graph Contrastive Learning (IDCL) model.
• Figure 3: Independence analysis on ML-100K with predefined 16 intents. Figure(a)-(b) shows the results in IDCL of behavior and user, respectively, i.e., the cosine similarity between the factors, the diagonal blocks indicates that different factors capture independent information. Figure(c) indicates the result of Variant A (IDCL w/o ICL), the confused high similarity emerge even across different factors.
• Figure 4: TSNE visualization of the learned intent embeddings on ML-100k. The behavior samples are divided into two disjoint subsets.
• Figure 5: Two users' behavior distributions over all intents for three movies.