Cheng, Xueqi, Yang, Catherine, Zhao, Yuying, Wang, Yu Emma, Karimi, Hamid, & Derr, Tyler. (2025). BTS: A Comprehensive Benchmark for Tie Strength Prediction. Proceedings of the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. https://doi.org/10.1145/3711896.3737441
The rapid growth of online social networks highlights the importance of understanding the varying strengths of online relationships. However, measuring tie strength (TS)—the closeness or intensity of a relationship—remains challenging due to the lack of clear ground-truth labels, different research approaches, and limited model performance in real-world settings. To help address these challenges, we introduce BTS, a comprehensive Benchmark for Tie Strength prediction, designed to provide a standardized foundation for evaluating and improving TS prediction methods.
Our contributions include: TS Pseudo-Label Techniques—we organize tie strength into seven standardized pseudo-labeling techniques based on existing research; TS Dataset Collection—we compile data from three representative social networks and analyze class distributions and correlations among the generated pseudo-labels; TS Pseudo-Label Evaluation Framework—we propose a standardized way to assess the quality of pseudo-labels based on tie resilience; and Benchmarking—we test existing tie strength prediction models using the BTS dataset, examining how different experimental settings, models, and evaluation criteria affect results.
From this work, we derive key insights to improve current methods and point to promising future directions. The BTS dataset collection, along with the curation codes and experimental scripts, is publicly available at: https://github.com/XueqiC/Awesome-Tie-Strength-Prediction
