Semantic Modelling of Document Focus-Time for Temporal Information Retrieval.

An accurate understanding of the temporal dynamics of Web content and user behaviors plays a crucial role during the interactive process between search engine and users. In this work, we focus on how to improve the retrieval performance via a better understanding of the time factor. On the one hand, we proposed a novel method to estimate the focus-time of documents leveraging their semantic information. On the other hand, we introduced query trend time for understanding the temporal intent underlying a search query based on Google Trend. Furthermore, we applied the proposed methods to two search scenarios: temporal information retrieval and temporal diversity retrieval. Our experimental results based on NTCIR Temporalia test collections show that: (1) Semantic information can be used to predict the temporal tendency of documents. (2) The semantic-based model works effectively even when few temporal expressions and entity names are available in documents. (3) The effectiveness of the estimated focus-time was comparable to that of the article’s publication time in relevance modelling, and thus, our method can be used as an alternative or supplementary tool when reliable publication dates are not available. (4) The trend time can improve the representation of temporal intents behind queries over query issue time.