FRel: A Freshness Language Model for Optimizing Real-Time Web Search

Abstract

An effective information retrieval system must satisfy different users search intentions expecting a variety of queries categories, comprising recency sensitive queries where fresh content is the major user’s requirement. However, using temporal features of documents to measure their freshness remains a hard task since these features may not be accurately represented in recent documents. In this paper, we propose a language model which estimates the topical relevance and freshness of documents with respect to real-time sensitive queries. In order to improve recency ranking, our approach models freshness by exploiting terms extracted from recently posted tweets topically relevant to each real-time sensitive query. In our experiments, we use these fresh terms to re-rank initial search results. Then, we compare our model with two baseline approaches which integrate temporal relevance in their language models. Our results show that there is a clear advantage of using microblogs platforms, such as Twitter, to extract fresh keywords.