A First Principles Approach to Trust-Based Recommendation Systems

2024-06-17

arXiv.org

AI Summary

This paper addresses the insufficient robustness of social recommendation systems against adversarial attacks. To mitigate this vulnerability, we propose a multi-source collaborative filtering framework that jointly leverages trust graphs, item ratings, and intra-item similarities. Methodologically, we design a weighted-averaging ensemble architecture compatible with arbitrary user similarity metrics; explicitly model tamper-resistant trust structures to enhance attack resilience; and incorporate item-similarity constraints to improve prediction consistency. Extensive experiments demonstrate that our approach significantly improves both recommendation robustness—particularly in low-rating regimes—and overall accuracy. Empirical results validate the inherent stability of trust graphs under data poisoning attacks. All source code is publicly released, establishing a reproducible, trustworthy paradigm for robust recommendation research.

Technology Category

• Machine Learning: Adversarial Learning & Robustness
• Computer Vision: Adversarial Attacks & Robustness
• Multiagent Systems: Adversarial Agents

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

This paper explores recommender systems in social networks which leverage information such as item rating, intra-item similarities, and trust graph. We demonstrate that item-rating information is more influential than other information types in a collaborative filtering approach. The trust graph-based approaches were found to be more robust to network adversarial attacks due to hard-to-manipulate trust structures. Intra-item information, although sub-optimal in isolation, enhances the consistency of predictions and lower-end performance when fused with other information forms. Additionally, the Weighted Average framework is introduced, enabling the construction of recommendation systems around any user-to-user similarity metric. All the codes are publicly available on GitHub.

Problem

• Develops trust-based recommender systems in social networks
• Analyzes influence of item ratings and trust graphs
• Introduces Weighted Average framework for user similarity

Innovation

• Leverages item-rating for collaborative filtering
• Uses trust graphs to resist network attacks
• Introduces Weighted Average framework