Graph Reasoning for Explainable Cold Start Recommendation

Jibril Frej,Marta Knezevic,Tanja Kaser

2024-06-12

Abstract:The cold start problem, where new users or items have no interaction history, remains a critical challenge in recommender systems (RS). A common solution involves using Knowledge Graphs (KG) to train entity embeddings or Graph Neural Networks (GNNs). Since KGs incorporate auxiliary data and not just user/item interactions, these methods can make relevant recommendations for cold users or items. Graph Reasoning (GR) methods, however, find paths from users to items to recommend using relations in the KG and, in the context of RS, have been used for interpretability. In this study, we propose GRECS: a framework for adapting GR to cold start recommendations. By utilizing explicit paths starting for users rather than relying only on entity embeddings, GRECS can find items corresponding to users' preferences by navigating the graph, even when limited information about users is available. Our experiments show that GRECS mitigates the cold start problem and outperforms competitive baselines across 5 standard datasets while being explainable. This study highlights the potential of GR for developing explainable recommender systems better suited for managing cold users and items.

Information Retrieval

What problem does this paper attempt to address?

The paper focuses on the cold start problem in recommendation systems, which refers to the challenge of new users or items without historical interaction data. Current solutions include using knowledge graph (KG) to train entity embeddings or graph neural networks (GNN). However, these methods often rely on the similarity between users and all items, making them vulnerable to popularity bias and potentially recommending popular items that are not in line with users' preferences. The paper proposes a framework called GRECS, which applies graph reasoning (GR) to cold start recommendation. It utilizes explicit paths starting from users to navigate in the KG and find items that match users' preferences, even with limited user information.

GRECS has the advantage of alleviating popularity bias and outperforming competing baselines on five benchmark datasets, while providing interpretable recommendation results. It integrates new users or items into the KG through non-interactive relationships, computes relevant embeddings, and does not require additional training. Experimental results show that GRECS performs well in strict cold start recommendation for users and items, requiring fewer related relationships and providing accurate recommendations, especially in situations with limited information.

The paper also compares other methods such as meta-learning, large-scale language model analysis of user preferences, and KG-based methods, pointing out the issues of popularity bias and lack of interpretability in these methods. As the first GR framework adapted for cold start recommendation, GRECS aims to provide more suitable recommendation strategies, particularly for managing cold start users and items in recommendation systems.