Probabilistic Retrieval Models for Time-Sensitive Queries

The intent of a query is sometimes related to the time issued. For example, a query Yahoo may refer to Yahoo! Maps in the morning and Yahoo! Games in the evening. Based on this observation, we extend the search intent to a larger scale: sense. We believe that each search action has its own sense, and this sense differs in different time. If search engines could distinguish such sense by time, then the ranking produced will match a user’s information need more precisely. In this work, we use click-through data to help the search engines find the sense of the search action in a specific time. We build three log-based probabilistic models to rank the search results by time. In addition, we propose ”Sense” to help these models to rank more precisely when the log data is unreliable. Experimental results shows our models rank better then the original ranking from AOL data.