Query-Based Automatic Training Set Selection for Microblog Retrieval

Typical pseudo-relevance feedback models assume that the first-pass documents are the most relevant and use those documents to select feedback terms for query expansion. In real applications, however, short documents, such as microblogs, may not have enough information about the searched topic, thus increasing the chance that irrelevant documents will be included in the initial set of retrieved documents. This situation reduces a feedback model's ability to capture information that is relevant to users' needs, which makes determining the best documents for relevant feedback without requiring extra effort from the user a critical challenge. In this paper, we propose an innovative mechanism to automatically select useful feedback documents using a topic modeling technique to improve the effectiveness of pseudo-relevance feedback models. The main idea behind the proposed model is to discover the latent topics in the top-ranked documents that allow for the exploitation of the correlation between terms in relevant topics. To capture discriminative terms for query expansion, we incorporated topical features into a relevance model that focuses on the temporal information in the selected set of documents. Experimental results on TREC 2011-2013 microblog datasets illustrate that the proposed model significantly outperforms all state-of-the-art baseline models.