Query performance prediction for microblog search

Highlights：

• In this work, we study query performance prediction (QPP) in the microblog search domain; a domain in which QPP is still understudied.

• Our study is considered the most comprehensive in this domain in terms of the number of predictors, retrieval models, and test collections used.

• We propose a set of predictors that generally outperform the state-of-the-art predictors across different retrieval models and with different collections.

• Our results show that using expanded queries in predicting the performance of query expansion retrieval models gives much better prediction quality than using the original unexpanded queries.

• We found that temporal predictors that ignore the fact that not all queries are temporal are not very effective. We also noticed that the prediction of a set of best-performing predictors is much more effective over temporal queries than non-temporal ones.

• The strong performance of our proposed predictors shows their high potential to be utilized to improve microblog search and other closely-related tasks such as tweet timeline generation.

摘要

•In this work, we study query performance prediction (QPP) in the microblog search domain; a domain in which QPP is still understudied.•Our study is considered the most comprehensive in this domain in terms of the number of predictors, retrieval models, and test collections used.•We propose a set of predictors that generally outperform the state-of-the-art predictors across different retrieval models and with different collections.•Our results show that using expanded queries in predicting the performance of query expansion retrieval models gives much better prediction quality than using the original unexpanded queries.•We found that temporal predictors that ignore the fact that not all queries are temporal are not very effective. We also noticed that the prediction of a set of best-performing predictors is much more effective over temporal queries than non-temporal ones.•The strong performance of our proposed predictors shows their high potential to be utilized to improve microblog search and other closely-related tasks such as tweet timeline generation.