Identifying Behavioural Intents in a Complex Search Process Management System

Abstract

Understanding search intents is essential for improving search results and query recommendations. Currently, most search intent identification methods work on plain form search logs consisting of queries and clicks. Meanwhile, new search tools are being introduced by the CHI/HCI communities, allowing users to perform more types of behaviors than querying and clicking, and generating rich form search logs with more information than that in plain form search logs. In this paper, we seek to identify search intents by considering various types of behaviors performed in a complex search process management system. We use random forest to identify search intents based on the behaviors performed. Experimental results show that search intents could be identified by analyzing search behaviors, and certain types of behaviors can be good indicators of search intents.