Document Priors Based On Time-Sensitive Social Signals

Abstract

Relevance estimation of a Web resource (document) can benefit from using social signals. In this paper, we propose a language model document prior exploiting temporal characteristics of social signals. We assume that a priori significance of a document depends on the date of users actions (social signals) and on the publication date (first occurrence) of the document. Particularly, rather than estimating the priors by simply counting signals related to the document, we bias this counting by taking into account the dates of the resource and the action. We evaluate our approach on IMDb dataset containing 167438 resources and their social data collected from several social networks. The experiments show the interest of temporally-aware signals at capturing relevant resources.