Details of Research Outputs

When a user performs a search on a search engine, the query reflects a particular interest of the user. The interest may either span a short session of a few minutes, or a long period of time like months or years. In the latter, the user may perform searching related to a particular interest from time to time, making the queries related to that interest sporadically distributed in the search log. Identification of these topically related queries is very meaningful, since it can help the search engine better understand the user's interest and hence deliver better results to the user. In this paper, we propose a method to aggregate topically related queries into interests regardless of where the queries appear in the search log. It first identifies sets of continuous topically-related queries called CTQs and then clusters similar CTQs together to form interests. In order to identify the CTQs accurately, we propose the Pattern-Concept-Time-Based (PCTB) method that utilizes query reformulation patterns, concepts behind the queries and the user's continuous searching behavior to compute the similarity between two queries. To evaluate the effectiveness of our approach, we employ the AOL search log as our test dataset and develop a search middleware on top of Google for extracting concepts related to the queries. Experimental results show that our method can obtain a high precision and recall on identifying CTQs, which in turn improves the performance of interest identification. © Springer-Verlag 2013.