Analysis of Twitter Lists as a Potential Source for Discovering Latent Characteristics of Users
Dongwoo Kim, Yohan Jo, Il‐Chul Moon, Alice H. Oh · 2010
This paper presents the results of a study using Twitter lists to infer the characteristics of users, especially about their interests. Gathering and structuring users ’ interest has been challenging because it often requires expen-sive data such as users ’ history logs or user surveys. Our approach overcomes this limitation and acquires infor-mative words representing user interests by using Twit-ter data which is open to the public and analyzing it with the χ2 feature selection algorithm. Furthermore, by using the tweets of all the users in a Twitter list, we can discover latent user characteristics that are not present in the tweets of individual users. We crawled Twitter list data that includes about ten percent of the Twitter user population, the lists they belong to, and the tweets of all the members of the lists. We used a stan-dard feature selection algorithm and a supervised clas-sification algorithm to verify the semantic coherence of the lists. Then we conducted a user survey which con-firmed that lists serve as good groupings of Twitter users with respect to the perceived characteristics of the users. The user survey also confirmed that the words extracted from each set are representative of all the members in the list whether or not the words are used explicitly by those members. 1.