A Time-Aware Language Model for Microblog Retrieval

Abstract : This paper describes our work (the IIEIR participation) in the TREC 2012 Microblog Adhoc Track. We proposed a ranking algorithm with temporal information based query. More and more research work proved that time is an important factor for improving the search result, especially for Microblog search. Based on Language Model, the representative work used time information as the document's prior information. Intuitively, there were two ways for making use of this feature. One way was query relevant while the other was query irrelevant. The hypothesis of the two models is the newer of the document, the more important. However, different query had different hot time points (the top time of relevance documents' time distribution). Take this into consideration; we supposed four models based on hot time points (HTLM). On this basis, we considered the model which is not relevant with query as document's background information and the model which is relevant with query as document's independent information. We used smoothing operation and supposed a mix timed language model. The results suggested that HTLM models are more effective for Microblog search and mix model further improved compared with the single model.