iTrace: An Implicit Trust Inference Method for Trust-aware Collaborative Filtering

Xu He, Bin Liu, Ke-Jia ChenarXiv 2017 · DOI · OpenReview

Abstract

The growth of Internet commerce has stimulated the use of collaborative filtering (CF) algorithms as recommender systems. A standard CF algorithm recommends items of interest to the target user by leveraging the votes given by other similar users, assuming that the credibility of every voting user is exactly the same. This assumption is often violated in practice, leading to misleading recommendations. To address this, we propose iTrace, a trust-aware CF (TaCF) algorithm that predicts implicit trust from a sparse explicit trust matrix. By leveraging both explicit and predicted implicit trust, iTrace significantly improves recommendation quality. Empirical evaluations on the Filmtrust dataset demonstrate that iTrace provides a substantial reduction in Mean Absolute Error (MAE) compared to traditional CF and existing trust-aware methods.

Introduction

Recommender systems are essential for navigating the vast choices in modern e-commerce. Collaborative Filtering (CF) is the most widely used technique, identifying "neighbors" with similar tastes to predict a target user's preferences. However, standard CF has a critical limitation: it ignores user credibility. In social contexts, recommendations from trusted friends are more valuable than those from strangers, even if tastes align.

Trust-aware Collaborative Filtering (TaCF) incorporates social trust to weight recommendations. The primary challenge for TaCF is data sparsity. Explicit trust statements (e.g., "User A trusts User B") are rare, leaving the trust matrix mostly empty. iTrace overcomes this by inferring implicit trust relationships across the network, effectively filling the gaps and mitigating the "cold start" problem for new users.

The iTrace Framework

The iTrace algorithm operates by integrating trust inference into the standard CF pipeline. It processes explicit trust data and rating data independently before fusing them to predict final ratings. This dual-path approach ensures that the system remains robust even when rating data is sparse.

System architecture of the iTrace trust-aware collaborative filtering framework.

Implicit Trust Inference

The core innovation of iTrace is its offline implicit trust inference module. It categorizes trust flow into four patterns: direct explicit trust, common trustees, trust propagation (transitivity), and no path. To handle propagation, iTrace treats the trust network as a weighted directed graph. It employs Dijkstra’s algorithm on a reciprocal trust matrix to find the most reliable path between users.

Crucially, iTrace models trust attenuation. It penalizes trust scores based on the length of the path and the number of intermediate nodes. This reflects the real-world intuition that trust degrades as it travels through a social chain. The resulting "Estimated Full Trust Matrix" is then fused with standard user similarity (Pearson correlation) to weight the rating predictor.

Experimental Evaluation

The algorithm was evaluated using the Filmtrust dataset, which contains 1,508 users, 2,071 items, and 35,416 ratings. The explicit trust matrix is highly sparse, with only 1,642 direct trust links. We compared several versions of iTrace against traditional CF and other state-of-the-art trust-aware methods using Mean Absolute Error (MAE) across various neighborhood sizes ($K$).

Key Results

The experimental results confirm that iTrace significantly outperforms traditional CF and existing TaCF baselines.

Comparison of MAE across different neighborhood sizes (K) for iTrace and baseline methods.

Key findings include:

  • Superior Accuracy: The optimal version, iTrace-I (using incremental weighting and attenuation-aware shortest paths), consistently achieved the lowest MAE.
  • The Value of Attenuation: iTrace variants that included path-length penalties outperformed those that did not, proving that modeling trust decay is essential for accurate inference.
  • Efficiency: Because the computationally intensive trust inference is performed offline, iTrace provides high-quality recommendations without sacrificing online response speed.

Conclusion

iTrace effectively addresses the sparsity problem in trust-aware recommender systems by inferring implicit trust through an attenuation-aware shortest-path approach. By fusing these inferred trust scores with traditional similarity metrics, iTrace provides more accurate and credible recommendations. Future work will explore more complex social interactions and temporal dynamics to further refine the trust model.

Summary updated

Abstract

The growth of Internet commerce has stimulated the use of collaborative filtering (CF) algorithms as recommender systems. A collaborative filtering (CF) algorithm recommends items of interest to the target user by leveraging the votes given by other similar users. In a standard CF framework, it is assumed that the credibility of every voting user is exactly the same with respect to the target user. This assumption is not satisfied and thus may lead to misleading recommendations in many practical applications. A natural countermeasure is to design a trust-aware CF (TaCF) algorithm, which can take account of the difference in the credibilities of the voting users when performing CF. To this end, this paper presents a trust inference approach, which can predict the implicit trust of the target user on every voting user from a sparse explicit trust matrix. Then an improved CF algorithm termed iTrace is proposed, which takes advantage of both the explicit and the predicted implicit trust to provide recommendations with the CF framework. An empirical evaluation on a public dataset demonstrates that the proposed algorithm provides a significant improvement in recommendation quality in terms of mean absolute error (MAE).
BibTeX
@misc{he2017itrace,
  author = {Xu He and Bin Liu and Ke-Jia Chen},
  title = {{iTrace: An Implicit Trust Inference Method for Trust-aware Collaborative Filtering}},
  year = {2017},
  eprint = {1708.03658},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1708.03658},
}
Figure