时间

2019年1月18号 11:00-12:00

 

地点

北京大学计算机所大楼106会议室

 

介绍

The gap between unstructured natural language and structureddata makes it challenging to build a system that supportsusing natural language to query large knowledge graphs.Many existing methods construct a structured query for theinput question based on a syntactic parser. Once the inputquestion is parsed incorrectly, a false structured querywill be generated, which may result in false or incompleteanswers. The problem gets worse especially for complex questions.In this paper, we propose a novel systematic methodto understand natural language questions by using a largenumber of binary templates rather than semantic parsers.As sufficient templates are critical in the procedure, we presenta low-cost approach that can build a huge number oftemplates automatically. To reduce the search space, wecarefully devise an index to facilitate the online templatedecomposition. Moreover, we design effective strategies toperform the two-level disambiguations (i.e., entity-level ambiguityand structure-level ambiguity) by considering thequery semantics. Extensive experiments over several benchmarksdemonstrate that our proposed approach is effectiveas it significantly outperforms state-of-the-art methods interms of both precision and recall. 

 

简历

Weiguo Zheng is an associate professor at the School of Data Science, Fudan University. He received his Ph.D. from Peking University in 2015. After his graduation, he worked as a postdoctoral research fellow at the Chinese University of Hong Kong. His research interest is graph data management and query processing. Currently, he is focusing onthe query and mining over knowledge graphs. He has published several papers in the top conferences and journals including SIGMOD、VLDB、ICDE、TODS、TKDE、CIKM.