Learning Knowledge from User Search

Learning Knowledge from User Search

In this paper, we introduce the concept of a novel application, called Knowledge Learning from User Search, aiming at identifying timely new knowledge triples from user search log. In the literature, the need of knowledge enrichment has been recognized as the key to the success of knowledge-based search. However, previous work of automatic knowledge extraction, such as Google Knowledge Vault, attempt to identify the unannotated knowledge triples from the full web-scale content in the offline execution. In our study, we show that most people demand a specific knowledge, such as the marriage between Brad Pitt and Angelina Jolie, soon after the information is announced. Moreover, the number of queries of such knowledge dramatically declines after a few days, meaning that the most people cannot obtain the precise knowledge from the execution of the offline knowledge enrichment. To remedy this, we propose the SCKE framework to extract new knowledge triples which can be executed in the online scenario. We model the'Query-Click Page'bipartite graph to extract the query correlation and to identify cohesive pairwise entities, finally statistically identifying the confident relation between entities. Our experimental studies show that new triples can also be identified in the very beginning after the event happens, enabling the capability to provide the up-to-date knowledge summary for most user queries.