Using graph evolutionary to retrieve related tweets

Abstract

Due to its popularity and daily active users, social media has become powerful and influential in the last decade. With the nature of a micro-blogging platform, instant messages and the latest short posts are sent throughout the network on Twitter. Therefore, most users utilize Twitter to update breaking news or the latest events. Since a huge volume of tweet messages have been published on Twitter, event evolution has also rapidly developed into related events within similar topics. In this study, we present a novel method to retrieve tweets that relate to a given query term. Not only perfectly matched tweets, but more related tweets will be retrieved. The collected tweet data are processed and constructed as an original network. With the benefits of social network analysis, a simplification-based summarization approach is applied to ignore information that has less importance while preserving significant information in the network based on centrality measurement and clustering coefficient. Using the evolutionary of graph-based representation extends the relationship diffusion to assist related information retrieval. Experiments were performed using Thai news datasets and the framework performance was evaluated by precision, recall, and f-score. The experimental results show that our framework outperformed the baseline methods which derived a similarity score based on the word-embedded vector to find relevant documents.