Leveraging Topic Features in Prediction of Social Network Community Evolutions

While the cornerstone of community evolution revolves around discussion topics, both the structure and topics undergo dynamic changes simultaneously. Most studies in this field have focused on complex network structural features, overlooking the influence of topics and their features on network complexity and time consumption. Topic features, in particular, play a critical role in shaping the community’s evolution and are increasingly recognized for their influence in driving changes within these digital ecosystems. Our study aims to fill this research gap by proposing topical features to enhance the accuracy of tracking and efficiency of predictive models. In this study, we developed the Identification of Community Evolution by Mapping “ICEM” method by integrating topic modeling with network structure analysis to track community evolution. We validated our approach using two distinct datasets from different linguistic contexts: Arabic and English social media communities, allowing for cross-linguistic validation of our findings. We employed three predictive classifiers to assess the significance of topical features and compared their performance against and in combination with structural features across both datasets. Our findings not only demonstrate the effectiveness of topical features in predicting community evolution with high accuracy across different languages, but also highlight significant variations in feature effectiveness between linguistic contexts, along with substantial reduction in execution time compared to existing approaches.