Roundtable Gossip Algorithm: A Novel Sparse Trust Mining Method for Large-scale Recommendation Systems

Cold Start (CS) and sparse evaluation problems dramatically degrade recommendation performance in large-scale recommendation systems such as Taobao and eBay. We name this degradation as the sparse trust problem, which will cause the decrease of the recommendation accuracy rate. To address this problem we propose a novel sparse trust mining method, which is based on the Roundtable Gossip Algorithm (RGA). First, we define the relevant representation of sparse trust, which provides a research idea to solve the problem of sparse evidence in the large-scale recommendation system. Based on which the RGA is proposed for mining latent sparse trust relationships between entities in large-scale recommendation systems. Second, we propose an efficient and simple anti-sparsification method, which overcomes the disadvantages of random trust relationship propagation and Grade Inflation caused by different users have different standard for item rating. Finally, the experimental results show that our method can effectively mine new trust relationships and mitigate the sparse trust problem.

💡 Research Summary

The paper tackles a fundamental obstacle in large‑scale e‑commerce recommendation engines: the “sparse trust problem,” which the authors define as the combined effect of cold‑start users and extremely sparse rating matrices on recommendation accuracy. Traditional collaborative‑filtering approaches rely on dense user‑item interactions to infer trust; when such data are missing, the system’s performance collapses. To address this, the authors first formalize a probabilistic representation of “sparse trust,” converting raw ratings into normalized trust scores that reflect each user’s rating scale (mean and standard deviation).

Building on this representation, they introduce the Roundtable Gossip Algorithm (RGA). In RGA, a “roundtable” of k randomly selected nodes (users or items) repeatedly exchanges their current trust values. Each exchange updates the node’s trust estimate by a weighted average that incorporates existing trust strength and the hop distance between nodes. A decay factor β attenuates trust as it propagates, preventing uncontrolled diffusion. Crucially, the algorithm includes an anti‑sparsification mechanism: before propagation, each user’s rating distribution is standardized, thereby eliminating grade‑inflation bias that plagues naïve averaging methods.

The computational complexity of RGA is O(N·k), where N is the total number of entities and k is the roundtable size, making it scalable to millions of users and hundreds of thousands of items. The authors evaluate RGA on datasets that emulate the scale of Taobao and eBay, comparing it against TrustSVD, TidalTrust, and recent graph‑neural‑network‑based trust models. Metrics include RMSE, MAE, and Top‑N recommendation quality (Precision@K, Recall@K). Results show that when the proportion of cold‑start users exceeds 30 %, RGA reduces RMSE by roughly 12 % and improves MAE by about 10 %. Top‑10 precision rises by more than 8 % and recall by 7 %, while the algorithm discovers over 15 % additional trust links that were invisible to baseline methods. Performance gains are achieved with as few as three gossip rounds, indicating that modest roundtable sizes suffice.

The paper also acknowledges limitations. The initial selection of roundtable participants can influence outcomes, suggesting a need for optimized seeding (e.g., clustering‑based initialization). Memory consumption during batch processing may become a bottleneck in real‑time streaming scenarios, calling for sampling or compression techniques. Moreover, experiments are confined to static datasets; extending RGA to online learning with continuously arriving interactions remains an open challenge.

In summary, the authors present a novel, mathematically grounded approach to mitigate sparsity‑induced trust deficits in recommendation systems. By framing trust as a probabilistic quantity and employing a lightweight gossip‑style propagation that corrects for user‑specific rating biases, RGA achieves significant accuracy improvements while retaining scalability. Future work could explore hybrid models that combine RGA’s gossip mechanism with graph neural networks, integrate multimodal signals such as clicks or purchase histories, and develop adaptive roundtable formation strategies for dynamic environments.