Abstract

Twitter and other social media provide the functionality of manually grouping users into lists. The goal is to enable se-lective viewing of content and easier information acquisition. However, creating lists manually requires significant time and effort. To mitigate this effort, a number of recent methods attempt to create lists automatically using content and/or net-work structure, but results are far from perfect. In this work, we study the power of the millions of lists that are already created by other twitter users in order to “crowdsource ” the task of list creation. We find that in a large dataset, collected specifically for this study, an opti-mal matching of existing lists from other twitter users to the ground-truth lists in egonets gives an F1 score of 0.43, while the best existing method achieves only 0.21. We explore the informativeness of features derived from net-work structure, existing lists, and posted content. We observe that different types of features are informative for different lists, and introduce a simple algorithm for ranking candidate lists. The proposed algorithm outperforms existing methods, but still falls short of the optimal selection of existing lists.