Knowledge-aware Coupled Graph Neural Network for Social Recommendation (2110.03987v1)

Abstract: Social recommendation task aims to predict users' preferences over items with the incorporation of social connections among users, so as to alleviate the sparse issue of collaborative filtering. While many recent efforts show the effectiveness of neural network-based social recommender systems, several important challenges have not been well addressed yet: (i) The majority of models only consider users' social connections, while ignoring the inter-dependent knowledge across items; (ii) Most of existing solutions are designed for singular type of user-item interactions, making them infeasible to capture the interaction heterogeneity; (iii) The dynamic nature of user-item interactions has been less explored in many social-aware recommendation techniques. To tackle the above challenges, this work proposes a Knowledge-aware Coupled Graph Neural Network (KCGN) that jointly injects the inter-dependent knowledge across items and users into the recommendation framework. KCGN enables the high-order user- and item-wise relation encoding by exploiting the mutual information for global graph structure awareness. Additionally, we further augment KCGN with the capability of capturing dynamic multi-typed user-item interactive patterns. Experimental studies on real-world datasets show the effectiveness of our method against many strong baselines in a variety of settings. Source codes are available at: https://github.com/xhcdream/KCGN.

Knowledge-aware Coupled Graph Neural Network for Social Recommendation

This paper addresses the problem of social recommendation by introducing a novel framework called Knowledge-aware Coupled Graph Neural Network (KCGN). The primary objective of social recommendation is to integrate users' social connections into preference prediction for items, aiming to improve the accuracy of recommendations, especially in scenarios where data sparsity is a significant issue. Existing models, primarily built on neural networks, have shown potential but are limited in three main aspects: 1) they often neglect inter-dependent knowledge across items, 2) they struggle with capturing interaction heterogeneity, and 3) they inadequately explore the dynamic nature of user-item interactions.

To overcome these challenges, KCGN is proposed, which incorporates inter-dependent knowledge across both items and users within a unified recommendation framework. This model facilitates high-order user- and item-wise relation encoding by leveraging mutual information maximization for enhanced global graph structure awareness. Furthermore, the model captures dynamic, multi-typed user-item interactive patterns and augments interactions within such patterns. The empirical results demonstrated through experimental studies on several real-world datasets ascertain the superiority of the proposed method over multiple strong baselines, ensuring its effectiveness across various settings.

The methodology of KCGN is characterized by several distinct features:

1. Multi-typed Interactive Pattern Modeling: It handles interaction heterogeneity by constructing a multi-typed interaction tensor, representing various types of user-item interactions. This allows for the differentiation of interaction types using a message passing framework, which incorporates both relation-aware and temporal encoding strategies.
2. Knowledge-aware Coupled Graph Neural Module: It combines local relational structure modeling with global graph substructure awareness, utilizing a mutual information maximization paradigm. This ensures the preservation of both node-specific user/item characteristics and global dependencies between graph elements.
3. Model Optimization: The optimization strategy integrates a pairwise BPR loss with mutual information maximization, balancing the interactions among different elements to comprehensively capture collaborative signals.

Empirical studies exhibit the robustness of KCGN, particularly with inactive user groups and cold-start scenarios, where user-item interactions are sparse. Furthermore, qualitative analyses on the embeddings derived through KCGN highlight its ability to distinctly cluster user-item pairs, demonstrating enhanced representation learning capabilities.

The practical implications of this research are substantial. The ability to integrate social information along with item relational knowledge aligns well with current trends in leveraging auxiliary data sources for improved recommendation quality. In theory, the synergy between graph neural network structures and mutual information estimation opens pathways for future developments in AI systems that demand high-order relational encoding.

In conclusion, KCGN fills crucial gaps in social recommendation systems by combining user and item interdependencies with dynamic interaction modeling. This comprehensive approach not only addresses present limitations but also sets the stage for future work that might explore even richer interaction types or more complex temporal dynamics in social networks.