Volunteer Future Collaboration Prediction with Topology-Based Ensemble Models

Volunteering efforts are important for achieving collective goals during crisis time. To facilitate such efforts, this study uses volunteer and task information to predict the likelihood of volunteers’ future collaboration with machine learning approaches. This study adopts data collected from the mobile crowdsourcing app “Anti-Pandemic Pioneers” used for organizing volunteers during the COVID-19 pandemic in Shenzhen, China. Volunteer information is constructed as an undirected, weighted graph, and this task can be considered as a link prediction problem in network theory. Multiple methodologies are experimented with for this problem, such as topology based-methods, node attribute-based methods, and mixed methods. This study also evaluates the performance of the node embedding approach. Finally, the experiment results show that the ensemble model of the Jaccard index, Adamic-Adar index, and resource allocation index outperforms all other approaches, achieves an accuracy score of 0.85333 for the private test, and ranks 2nd on the task2 leaderboard.