A clustering-based framework for efficient and fair data valuation in federated trajectory systems

Existing research has primarily focused on utilizing shared trajectory data; however, in real-world applications, trajectory data is typically held by multiple distinct entities, forming a data federation. Constrained by the lack of effective incentive mechanisms, data owners are often unwilling to share their data without receiving fair value in return, which leads to data silo issues. Traditional data valuation methods struggle to fairly and efficiently quantify the contributions of each party in such federated environments. To address this problem, Professor Yongxin Tong's team from Beihang University published their latest research findings in "Frontiers of Computer Science" (co-published by Higher Education Press and Springer Nature). The team proposed a clustering-based efficient valuation framework for federated trajectory data, aiming to efficiently quantify the value of each party's data to incentivize data sharing, while ensuring the fairness of the valuation. Experiments on real datasets demonstrate that the proposed framework outperforms baseline methods in terms of both computational speed and valuation accuracy.

The framework first leverages the decomposable property of the total utility function over "road segment-time" pairs, transforming the complex global valuation problem into the assessment of numerous independent units. Subsequently, the framework groups "road segment-time" pairs that exhibit similar trajectory coverage patterns using a clustering algorithm, thereby identifying redundancies in computation. Finally, by computing the Shapley value only for the center of each cluster and propagating this result to all members within that cluster, the framework significantly reduces redundant computations and efficiently obtains the total value of each dataset. This process notably enhances the efficiency and accuracy of data value assessment while ensuring fairness.