Expertise-Aware Crowdsourcing Taxonomy Enrichment.

WISE(2021)

引用 0|浏览32
暂无评分
摘要
Taxonomy is an indispensable part of Knowledge Base. Taxonomy Enrichment is the task that new instances are classified to possibly thousands of concepts of an existing taxonomy. Recent works considered to enrich a taxonomy by crowdsourcing [17,24] to promote the quality of new instance classification. However, the broad coverage of knowledge induced by a large taxonomy makes it a skill-sensitive knowledge-intensive crowdsourcing (KI-C) task. Naively applying existing skill estimation methods [23,25,36] for quality control requires quizzes on each concept of the taxonomy, which is prohibitive for large scale Taxonomy Enrichment. In this work, we propose a unified crowdsourcing framework to mitigate both challenges. It leverages the skill locality of workers with a Graph Gaussian Process model. Our quality control method automatically matches instances with a worker's expertise and gives hints for later workers to reduce their time to search over the taxonomy. In our experiments with 314 workers, our method outperforms baseline BTSK [9] by 6.6% in terms of accuracy on a taxonomy of 10 levels and 632 nodes.
更多
查看译文
关键词
Crowdsourcing,Quality control,Taxonomy construction,Taxonomy enrichment,Knowledge base
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要