Personalizing information retrieval using task features, topic knowledge, and task product

• Authors: Jingjing Liu
• Venue: Proceedings of the 32nd international ACM SIGIR conference on Research and development in information retrieval
• Year: 2009

Personalization of information retrieval tailors search towards individual users to meet their particular information needs. Personalization systems obtain additional information about users and their contexts beyond the queries they submit to the systems, and use this information to bring the desired documents to top ranks. The additional information can come from various sources: user preferences, user behaviors, contexts, etc. [1] To avoid users taking extra effort in providing explicit preferences, most personalization approaches have adopted an implicit strategy to obtain users' interests from their behaviors and/or contexts, such as query history, browsing history, and so on. Task, topic knowledge, and desktop information have been used as evidence for personalization. Tailoring display time threshold based on task information was found to improve implicit relevance feedback performance [5]. User's familiarity with search topics was found to be positively correlated with reading time but negatively correlated with search efficacy [3]. This indicated the possibility of inferring topic familiarity from searching behavior. Desktop information was also found to be a good source for personalization [2, 4], and personalization using only those files relevant to user queries are more effective than using the entire desktop data [2]. Since search often happens in a work task environment, we examine how user-generated products and retained documents can help improve search performance. To these ends, this study looks at how the following factors can help personalize search: features of user's work tasks (including task stage and task type), user's familiarity with work task topic, user's saving and using behaviors, and task product(s) that the user generated for the work task. Work tasks are designed to include multiple sub-tasks, each being a stage. Two types of sub-task interdependence are considered: parallel, where the sub-tasks do not depend upon each other, or dependent, where one sub-task depends upon the accomplishment of other sub-task(s). The study examines the interaction effects of these factors, dwell time, and document usefulness. It also looks at a personalization technique that extracts terms for query expansion from work task product(s) and user behaviors. There are three research questions: RQ1: Does the stage of the user's task help predict document usefulness from dwell time in the parallel and the dependent tasks, respectively? RQ2. Does the user's familiarity with work task topic help predict document usefulness from dwell time in the parallel and the dependent tasks, respectively? RQ3. Do user's task product(s) and saving and using behaviors help with query disambiguation? Twenty-four participants are recruited, each coming three times (as three experiment sessions) to a usability laboratory working on three sub-tasks in a general task, either a parallel or a dependent. Take the parallel task as an example. It asks the participants to write a three-section article on hybrid cars, and each section is finished in one session. The three sections focus on Honda Civic sedan hybrid, Nissan Altima sedan hybrid, and Toyota Camry sedan hybrid, respectively. When searching for information, half of the participants use a query expansion condition, where the system recommends search terms based on their work in previous sessions, and the other half use a non-query expansion system condition. Data are collected by three major means: logging software that records user-system interactions, an eye tracker that records eye movement, and questionnaires that elicit users' background information and their perceptions on a number of aspects. The results will provide new evidence on personalizing search by taking account of the examined contextual factors.