Date of Award

Fall 2015

Document Type


Degree Name

Doctor of Philosophy in Information Systems - (Ph.D.)


Information Systems

First Advisor

Yi-Fang Brook Wu

Second Advisor

Yi Chen

Third Advisor

Zhi Wei

Fourth Advisor

Songhua Xu

Fifth Advisor

Lian Duan


The information needs of search engine users vary in complexity. Some simple needs can be satisfied by using a single query, while complicated ones require a series of queries spanning a period of time. A search task, consisting of a sequence of search queries serving the same information need, can be treated as an atomic unit for modeling user’s search preferences and has been applied in improving the accuracy of search results. However, existing studies on user search tasks mainly focus on applying user’s interests in re-ranking search results. Only few studies have examined the effects of utilizing search tasks to assist users in obtaining effective queries. Moreover, fewer existing studies have examined the dynamic characteristics of user’s search interests within a search task. Furthermore, even fewer studies have examined approaches to selective personalization for candidate refined queries that are expected to benefit from its application. This study proposes a framework of modeling user’s task-based dynamic search interests to address these issues and makes the following contributions. First, task identification: a cross-session based method is proposed to discover tasks by modeling the best-link structure of queries, based on the commonly shared clicked results. A graph-based representation method is introduced to improve the effectiveness of link prediction in a query sequence. Second, dynamic task-level search interest representation: a four-tuple user profiling model is introduced to represent long- and short-term user interests extracted from search tasks and sessions. It models user’s interests at the task level to re-rank candidate queries through modules of task identification and update. Third, selective personalization: a two-step personalization algorithm is proposed to improve the rankings of candidate queries for query refinement by assessing the task dependency via exploiting a latent task space. Experimental results show that the proposed TOQUE framework contributes to an increased precision of candidate queries and thus shortened search sessions.