Query Expansion and Message-Passing Algorithms for TREC Microblog Track

This report describes the methods that our Information Retrieval Group at Purdue University used for the TREC Microblog 2011 track. The first method is the pseudo-relevance feedback, a traditional algorithm to reformulate the query by adding expanded terms to the query. The second method is the affinity propagation, a non parametric clustering algorithm that can group the top tweets according to their similarities. The final score of a tweet is based on its relevance score and the relevance score of its representative in the group. We found that query expansion is a very useful technique for microblog retrieval, while affinity propagation could achieve a comparable performance when combining with other techniques.