Reconstruction and Analysis of Twitter Conversation Graphs

Communication over social networks has been an emergent theme over the last several years. We wish to examine the underlying graph structure in social networks and to understand the dynamics of how information propagates in these networks. Gathering data from large social networks such as Facebook and Twitter has become increasingly more difficult due to external pressure to improve privacy and internal pressure to generate revenue. We examine the case of Twitter, the micro-blogging social network. Twitter provides APIs which allow users to sample tweets stating specific keywords - however the reconstruction of complete conversations (which is integral to the dynamics of how information propagates) is difficult. We focus on the particular case of a conversation formed by users replying to tweets. We present a robust approach for the reconstruction of complete conversations and compare the resultant graph measurements to those created using keyword searches.