Abstract
A variety of rating-based recommendation methods have been extensively studied including the well-known collaborative filtering approaches and some network diffusion-based methods, however, social trust relations are not sufficiently considered when making recommendations. In this paper, we contribute to the literature by proposing a trust-based recommendation method, named CosRA+T, after integrating the information of trust relations into the resource-redistribution process. Specifically, a tunable parameter is used to scale the resources received by trusted users before the redistribution back to the objects. Interestingly, we find an optimal scaling parameter for the proposed CosRA+T method to achieve its best recommendation accuracy, and the optimal value seems to be universal under several evaluation metrics across different datasets. Moreover, results of extensive experiments on the two real-world rating datasets with trust relations, Epinions and FriendFeed, suggest that CosRA+T has a remarkable improvement in overall accuracy, diversity, and novelty. Our work takes a step towards designing better recommendation algorithms by employing multiple resources of social network information.
Introduction
Traditional recommendation algorithms primarily utilize Collaborative Filtering (CF) or Network Diffusion processes. CF focuses on identifying patterns in user-item interactions, while Diffusion models treat recommendations as a physical flow of resources through a bipartite network. However, these methods often overlook the critical role of social trust. In practice, users are more likely to trust recommendations from their social circle. This research introduces CosRA+T, a novel algorithm that integrates social trust relations into a network diffusion-based framework to improve recommendation accuracy, diversity, and novelty.
Methodology
CosRA+T builds upon the CosRA index, which merges cosine similarity with resource-allocation logic. The algorithm operates in three distinct phases:
- Initialization: A target user’s resources are distributed across the objects they have previously interacted with.
- Diffusion: Resources flow from these objects to all users who have collected them, identifying "neighbors" in the network.
- Redistribution with Trust Scaling: Before resources flow back to objects to generate the final recommendation list, the resources held by users whom the target user trusts are adjusted using a tunable scaling parameter $\theta$.
Mathematically, for a trusted user, their resource $f_j$ is transformed into $f_j^\theta$. Since these values are fractions between 0 and 1, setting $\theta < 1$ effectively boosts the influence of trusted peers in the final recommendation output.
Results and Discussion
The performance of CosRA+T was evaluated using two real-world datasets, Epinions and FriendFeed, which contain explicit user-user trust links. A significant finding of this study is the existence of an optimal value for the scaling parameter $\theta$.
As illustrated in the experimental results, ignoring trust ($\theta = 1$) results in lower accuracy, while over-weighting trust (as $\theta$ approaches 0) can lead to an "echo chamber" effect that also degrades performance. The optimal $\theta$ was found to be remarkably consistent across different evaluation metrics (AUC, Precision, Recall) and datasets, typically hovering between 0.65 and 0.70. This suggests a potentially universal pattern in the weight of trust in human decision-making. Furthermore, CosRA+T outperformed standard benchmarks such as User-based CF, Heat Conduction, and Mass Diffusion in terms of accuracy and provided a superior balance of diversity and novelty.
Conclusion
The CosRA+T method successfully demonstrates that incorporating social trust into network diffusion models can significantly enhance recommender system performance. The stability of the optimal scaling parameter across datasets highlights its robustness. Future work may involve extending the model to handle multi-level ratings instead of binary links and integrating community-wide user reputation to further refine the trust-based diffusion process.
Summary updated