Demonstration system for contrastive learning-based semi-supervised community search

Abstract

Community search, which aims to retrieve important communities for a given query vertex has substantial practical implications in network analysis. This significance is underscored by the fact that each vertex within these networks is tagged with a unique influence value, reflecting its relative importance or impact. The introduction of the COCLEP model marks a significant advancement in this field. Rooted in the principles of contrastive learning and incorporating partitioning strategies, the model is an efficient means of community search that only requires a few labels. In this project, we explore the user scenarios of the model, developing an interactive application designed to facilitate the visualization of COCLEP's outputs in comparison with the ground truth. The application should allow users to gain an in-depth, holistic comprehension of how COCLEP operates within the broader context of network analysis. This deepened understanding is crucial for users to effectively leverage the model in various practical scenarios, thereby enhancing the application of community search in real-world networks.