Exploring Pseudo-Relevance Feedback for Microblog Search

This study explored the effectiveness of a classical information retrieval (IR) approach, pseudo-relevance feedback (PRF), on improving the performance of microblog search. Factors including number of PRF iterations, term selection strategy, term weighting scheme and use of user-generated metadata were examined in order to shed light on their influence on the effectiveness of the studied approach in an environment of microblog search. An IR system implementing the studied approach was developed for experiments and experiments were conducted on an English microblog corpus composed of Twitter's microblogs, known as tweets. Search performance was evaluated using precision at thirty (P@30), mean average precision (MAP) and normalized discounted cumulative gain at thirty (NDCG@30). As a result, it was found that pseudo-relevance feedback can significantly improve performance of microblog search. Meanwhile, it was also revealed that expanding queries with hashtags is detrimental to the search performance. Besides, it was also identified that term weighting can contribute to search performance.Master of Science in Information Scienc