Query performance prediction for microblog search

Abstract

Query performance prediction (QPP) is the task of estimating the effectiveness of a retrieval system given a search query in the absence of any feedback from the searcher. The task has been proven to be very challenging, and thus it attracted a lot of research attention in domains like news and Web retrieval. However, search in microblogs poses new challenges for the task due to the more prevalent temporality in microblogs and the different types of information needs in such domain. In this work, we aim at studying QPP for microblog search. We conducted large-scale experiments, testing 37 state-of-the-art predictors using several types of retrieval models usually used in microblog search. Moreover, we propose a set of predictors that exhibit statistically-significant improvements over the state-of-the-art predictors with the maximum percentage of improvement reaching 55% over all studied retrieval settings. Further experimental explorations show that using expanded queries in predicting the performance of query expansion models gives much better prediction quality than using the original queries, and that the examined predictors were generally much more effective over temporal queries compared to non-temporal ones; both phenomena have never been studied in the context of microblog search before. As microblog search is considered a major step in several retrieval tasks in the domain (such as timeline generation, summarization, and question answering), improving QPP for microblog search has a high potential to help improve the effectiveness of those closely-related tasks. 1 2017 Elsevier Ltd