Tailor knowledge graph for query understanding: linking intent topics by propagation

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1070–1080, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Tailor knowledge graph for query understanding: linking intent topics by propagation Shi Zhao z.s@pku.edu.cn Yan Zhang zhy@cis.pku.edu.cn Department of Machine Intelligence, Peking University, Beijing, China Key Laboratory on Machine Perception, Ministry of Education, Beijing, China Abstract Knowledge graphs are recently used for enriching query representations in an entity-aware way for the rich facts or- ganized around entities in it. How- ever, few of the methods pay attention to non-entity words and clicked websites in queries, which also help conveying user intent. In this paper, we tackle the prob- lem of intent understanding with innova- tively representing entity words, refiners and clicked urls as intent topics in a uni- fied knowledge graph based framework, in a way to exploit and expand knowl- edge graph which we call ‘tailor’. We collaboratively exploit global knowledge in knowledge graphs and local contexts in query log to initialize intent representa- tion, then propagate the enriched features in a graph consisting of intent topics us- ing an unsupervised algorithm. The ex- periments prove intent topics with knowl- edge graph enriched features significantly enhance intent understanding. 1 Introduction Query understanding is the process of generating a representation which characterizes a user’s search intent (Croft et al., 2010), which is of vital im- portance for information retrieval. However, users are remarkably laconic in describing their infor- mation needs due to anomalous state of knowledge (Belkin et al., 1982), resulting in vague and under- specified queries, which makes it especially dif- ficult to understand and locate what they intended for in mountains of web data. The problem is often significantly compounded that people convey their intent rather in a series of behaviors called a search session than a single query, leaving a wealth of clues including query reformulations, page visits, dwell times, etc. What’s more, as entities are tak- ing center stage (Yin and Shah, 2010), string-level or phrase-level modeling of intent soon hits the bottleneck, calling for an entity-aware perspective. Knowledge repositories, better known as knowledge graphs, such as Wikipedia, DBpedia and Freebase, have been recently utilized for en- hancing query understanding for the large amounts of world knowledge they’ve harvested about en- tities and facts. A widely accepted way to use knowledge graph is tying queries with it by anno- tating entities in them, also known as entity link- ing. However, information need is conveyed through more than entities. Quite a few non-entity words, aka refiners or modifiers, as well as many urls are barely included in knowledge graph, while they play an irreplaceable role in intent understand- ing. For example, a user may query toyota, volvo or just enter car soup, cars for sale and click www.carsoup.com, which should be encoded in a form that we could perceive their closeness in intent. That’s why at-a-glance info cards about merely recognized entity in the query are far from enough and previous methods …

Content truncated to 500 words