Deconstructing Complex Search Tasks: a Bayesian Nonparametric Approach for Extracting Sub-tasks

Proceedings of NAACL-HLT 2016, pages 599–605, San Diego, California, June 12-17, 2016. c⃝2016 Association for Computational Linguistics Deconstructing Complex Search Tasks: A Bayesian Nonparametric Approach for Extracting Sub-tasks Rishabh Mehrotra Dept. of Computer Science University College London London, UK R.Mehrotra@cs.ucl.ac.uk Prasanta Bhattacharya Dept. of Information Systems National University of Singapore Singapore prasanta@comp.nus.edu.sg Emine Yilmaz Dept. of Computer Science University College London London, UK emine.yilmaz@ucl.ac.uk Abstract Search tasks, comprising a series of search queries serving a common informational need, have steadily emerged as accurate units for developing the next generation of task-aware web search systems. Most prior research in this area has focused on segmenting chrono- logically ordered search queries into higher level tasks. A more naturalistic viewpoint would involve treating query logs as convo- luted structures of tasks-subtasks, with com- plex search tasks being decomposed into more focused sub-tasks. In this work, we focus on extracting sub-tasks from a given collec- tion of on-task search queries. We jointly leverage insights from Bayesian nonparamet- rics and word embeddings to identify and ex- tract sub-tasks from a given collection of on- task queries. Our proposed model can inform the design of the next generation of task-based search systems that leverage user’s task behav- ior for better support and personalization. 1 Introduction Search behavior, and information behavior more generally, is often motivated by tasks that prompt search processes that are often lengthy, iterative, and intermittent, and are characterized by distinct stages, shifting goals and multitasking (Kelly et al., 2013; Mehrotra et al., 2016). Current search sys- tems do not provide adequate support for users tack- ling complex tasks, due to which the cognitive bur- den of keeping track of such tasks is placed on the searcher. Ideally, a search engine should be able to understand the reason that caused the user to submit a query (i.e., the actual task that caused the query to be issued) and be able to guide users to achieve their tasks by incorporating this information about the actual informational need. Clearly, identifying and analyzing search tasks is an extremely impor- tant activity not only for search engine providers but also other web based frameworks like spoken dia- logue (Sun et al., 2015) and general recommenda- tion systems (Mehrotra et al., 2014) in their effort to improve user experience on their platforms. Previous work in the area have proposed a num- ber of methods for identifying and extracting task knowledge from search query sessions (Mehrotra and Yilmaz, 2015b; Wang et al., 2013; Lucchese et al., 2011; Verma and Yilmaz, 2014; Mehrotra and Yilmaz, 2015a). However, while some tasks are fairly trivial and single-shot (e.g. ”latest Taylor Swift album”), others are more complex and often involve multiple steps or sub-tasks (e.g. ”planning a wedding”). Deciphering sub-tasks from search query logs be- comes an important problem since users might ex- hibit different search preferences as well as expend different amounts of search effort while executing the sub-tasks. For example, while planning a wed- ding, users might choose to spend more time …

Content truncated to 500 words