Topic-model based query intent prediction for search by multiple examples

It is often difficult for users to express their information needs as keywords. Search-By-Multiple-Examples (SBME) is a new search paradigm that provides an easier way for users to express their information needs as multiple relevant examples (called query examples) rather than as a simple string of keywords. One key issue of SBME is to identify user’s true information needs from the query examples, which may contain multiple topics. However, none of the previous studies on SBME considers the issue of topic diversity of the query examples. In this research, we explore the solutions to the topic diversity issue in SBME through adopting topic modeling techniques to predict the likelihood that the query examples belonging to a topic. The learned topic distributions from the query examples are used to build query vectors for document ranking. Using Mean Average Precision (MAP) and precision at k (p@k), experiments conducted on two benchmark datasets show that the proposed method outperforms the two baselines significantly, especially when a large number of query examples are available.