Volunteer Retention and Future Collaboration Prediction in Volunteer Crowdsourcing Platforms

This paper presents a unique challenge aimed at learning models that capture volunteers’ task participation and collaboration behaviors, using data collected from a mobile volunteer crowdsourcing platform. The dataset utilized for this challenge comprises volunteer activity records of 68,682 users during the period 2020-2022 in Shenzhen, China. The challenge is designed to address two specific tasks: first, predicting volunteers’ retention based on their historical participation, and secondly, forecasting future collaborations between pairs of volunteers. A total of 16 and 9 teams participated in the two tasks respectively, submitting their results on Kaggle for evaluation. The selection of winners was based on comprehensive considerations of both model performance and paper quality. The top two performing teams in each task were recognized as the winners, with the first-place teams achieving an impressive root mean square error (RMSE) of 76.37 in the first task and an Area under the Receiver Operating Characteristic Curve (AUCROC) of 0.8557 in the other. The challenge result demonstrates the potential of machine learning in predicting user behaviors in crowdsourcing volunteer platforms and draws insights into the key factors affecting volunteers’ retention and collaboration preferences.