Temporal Smoothing: Discriminatively Incorporating Various Temporal Profiles of Queries

Pengming Wang, Qing Chen, Bin Wang

Abstract: Document smoothing has been shown to play a critical role to deal with the zero probability problems in the query likelihood retrieval model. Unlike traditional approaches using the same corpus language model for every document, the mainstream of the current methods introduce an additional smoothing item that can reflect the content of each document. However, these methods would either ignore the temporal characteristics of queries, or handle temporal factors in a multi-step process that cannot be explained in an unified solution, which rules out many potentially good alternatives. We instead propose a novel method, called temporal smoothing, which can alleviate the above problems. In particular, by using the overall temporal distribution of documents to smooth the temporal profile of a given query, the estimated temporal query model can be used in query likelihood retrieval model as an unified solution. Empirical evaluations based on a collection of Twitter documents and some standard benchmarks demonstrate the effectiveness of the proposed temporal smoothing mechanism in the retrieval task.