不确定知识图谱中面向结构查询的众包清洗研究

项目名称： 不确定知识图谱中面向结构查询的众包清洗研究

项目编号： No.61572193

项目类型： 面上项目

立项/批准年度： 2016

项目学科： 自动化技术、计算机技术

项目作者： 林欣

作者单位： 华东师范大学

项目金额： 16万元

中文摘要： 知识图谱以图数据记录实体与实体间关系，为下一代搜索引擎提供语义级别支持。自动从互联网挖掘实体与关系是知识图谱构建的重要方式，然而由于自然语言处理和人工智能等技术的限制，自动挖掘出的关系可信度不高，从而导致在知识图谱之上的查询存在不确定性，质量不高，严重影响知识图谱的可用性。本项目拟采用众包等人工干预手段对知识图谱中不确定性关系进行清洗，以提高热门查询的结果质量。针对知识图谱中典型的结构查询：可达性查询，研究如何在不确定知识图谱中选取最合适的关系，以及如何对这些关系进行清洗，以达到结构查询结果质量提升的最大化。具体来说，我们研究：1）单查询单关系众包清洗；2）单查询多关系众包清洗；3)多查询多关系众包清洗。相信本项目的研究成果将很好地提升知识图谱的可用性，为其大规模推广扫清障碍。

中文关键词： 知识图谱；众包；可达性查询；；

英文摘要： Knowledge graph stores entities and relations in the form of graph, which can provide semantic supports for the next generation search engines. An important way to construct knowledge graph is automatically extracting entities and relations from Internet. Due to the limitation of natural language processing and artificial intelligence, the confidences of some automatically extracted relations are not high, which will degrade the quality of query results on the knowledge graphs. As such, the usability of the knowledge graphs is also severely degraded. In this project, we aim to promote the quality of query results by exploiting human-powered solution, such as crowdsourcing, to clean the uncertain relations in the knowledge graph. We focus on a typical structural query on knowledge graph, i.e., reachability query. We aim at researching how to properly select relations and how to clean these relations to maximizing the quality improvement. In detail, we will study 1) single-query single-relation crowdsourcing cleaning, 2) single-query multi-relation crowdsourcing cleaning and 3) multi-query multi-relation crowdsourcing cleaning. The result of this project will help to promote the usability of the knowledge graphs and clear the way for the large-scale promotion of them.

英文关键词： knowledge graph；crowdsourcing；reachability query；；