Temporal Relevance Profiles for Tweet Search

When searching tweets, users may know something about the temporal characteristics of the information they’re after. For example, based on external knowledge, a searcher might prefer more recent results or results within a particular time interval. However, most search applications do not allow the user to explicitly supply this information, and neither do most retrieval models have a mechanism to incorporate this additional evidence. In this paper, we introduce the notion of a temporal relevance profile, which a user explicitly includes alongside a keyword search query. We propose alternative representations of temporal relevance profiles and how existing retrieval models might take advantage of this data. Oracle experiments on microblog track data from TREC 2011 and 2012 empirically demonstrate that this approach has the potential to significantly increase the quality of retrieved results.