Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
102 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Explainable Reasoning over Knowledge Graphs for Recommendation (1811.04540v1)

Published 12 Nov 2018 in cs.IR

Abstract: Incorporating knowledge graph into recommender systems has attracted increasing attention in recent years. By exploring the interlinks within a knowledge graph, the connectivity between users and items can be discovered as paths, which provide rich and complementary information to user-item interactions. Such connectivity not only reveals the semantics of entities and relations, but also helps to comprehend a user's interest. However, existing efforts have not fully explored this connectivity to infer user preferences, especially in terms of modeling the sequential dependencies within and holistic semantics of a path. In this paper, we contribute a new model named Knowledge-aware Path Recurrent Network (KPRN) to exploit knowledge graph for recommendation. KPRN can generate path representations by composing the semantics of both entities and relations. By leveraging the sequential dependencies within a path, we allow effective reasoning on paths to infer the underlying rationale of a user-item interaction. Furthermore, we design a new weighted pooling operation to discriminate the strengths of different paths in connecting a user with an item, endowing our model with a certain level of explainability. We conduct extensive experiments on two datasets about movie and music, demonstrating significant improvements over state-of-the-art solutions Collaborative Knowledge Base Embedding and Neural Factorization Machine.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (6)
  1. Xiang Wang (279 papers)
  2. Dingxian Wang (14 papers)
  3. Canran Xu (12 papers)
  4. Xiangnan He (200 papers)
  5. Yixin Cao (138 papers)
  6. Tat-Seng Chua (360 papers)
Citations (416)

Summary

Explainable Reasoning over Knowledge Graphs for Recommendation

The paper "Explainable Reasoning over Knowledge Graphs for Recommendation" presents a model named Knowledge-aware Path Recurrent Network (KPRN) designed to integrate knowledge graphs (KGs) into recommender systems. The integration of KGs provides a rich auxiliary source of information, enhancing user-item interaction data by exploring interlinks between entities. KPRN aims to utilize this connectivity to infer user preferences by leveraging sequential dependencies among paths within the KG, thus enabling both recommendation accuracy and explainability.

Model Overview

The KPRN architecture focuses on generating path representations that consist of both entities and relations. This is accomplished by modeling the sequential dependencies within these paths, allowing the system to deduce the underlying rationale for user-item interactions. Notably, the model includes a weighted pooling mechanism that evaluates the strength of different paths' contributions to the recommendation process, adding a level of explainability to its predictions.

The model architecture comprises three main components:

  1. Embedding Layer: It projects the entity, entity type, and relation into a latent space. This process is crucial for capturing the semantic information of each path element.
  2. LSTM Layer: This recurrent network encodes the sequential information of path elements to infer the comprehensive semantics for the user-item interaction, leveraging the long-term dependencies inherent in the paths.
  3. Weighted Pooling Layer: This layer discriminates and aggregates the predictive strengths of multiple paths, facilitating path-wise explanations for recommendations.

Experimental Evaluation

The authors conducted extensive experiments using datasets from the movie (MovieLens-1M and IMDb) and music (KKBox) domains. These experiments demonstrated the superlative performance of KPRN over state-of-the-art models such as Collaborative Knowledge Base Embedding (CKE) and Neural Factorization Machine (NFM), underscoring the benefits of incorporating KG information, particularly when utilized explicitly through path-based reasoning.

Key results include:

  • KPRN consistently outperformed CKE, a prominent embedding-based method, highlighting the efficacy of explicit reasoning over implicit representation learning.
  • The use of multiple qualitative paths with relations distinctively improved model performance, emphasizing the importance of nuanced semantic path representations.
  • The weighted pooling operation offered superior adaptability in differentiating path contributions relative to simpler aggregation mechanisms like mean-pooling and max-pooling.

Implications and Future Directions

The implications of this research are manifold. Practically, the integration of KGs through KPRN can significantly enhance recommendation systems, particularly in scenarios of sparse interaction data. Theoretically, this work contributes to the growing interest in interpretable AI models by demonstrating how knowledge-enhanced frameworks can provide insights into system recommendations.

Looking ahead, the authors propose directions such as deploying Graph Neural Networks to avoid the laborious path extraction process, and exploring zero-shot learning applications across domain-overlapping KGs to handle cold start problems. These advancements could further improve the scalability and generalization capabilities of KG-enhanced recommender systems, steering towards more robust and insightful AI-driven solutions.