Tweetement: Pseudo-relevance Feedback for Twitter Search

Microblogging platforms such as Twitter let users communicate with short messages. Due to the messages’ short content and the users’ tendency to type short queries while searching, it is particularly challenging to locate useful tweets that match user queries. The fundamental problems of word mismatch due to ambiguity are especially acute. To solve this problem, this thesis explores and compares multiple automatic query expansion methods that involve the most frequent hashtags and keywords. We built a Web service that provides real-time Twitter Search results incorporating automatic query expansion. Six pseudo-relevance feedback methods were studied and the numbers indicate that results without query expansion perform just as well as results with query expansion. However, the expanded queries find different relevant tweets than the original query, indicating, from multiple methods, that combining the results is a fruitful area for future investigations. Keywords: microblog, Twitter Search, query expansion, pseudo-relevance feedback, Web servic