无指导的中文开放式实体关系抽取

Unsupervised Chinese Open Entity Relation Extraction

传统的实体关系抽取需要预先定义关系类型体系,然而定义一个全面的实体关系类型体系是很困难的.开放式实体关系抽取技术解决了预先定义关系类型体系的问题,但是在中文上的研究还比较少.提出面向大规模网络文本的无指导开放式中文实体关系抽取方法,首先使用实体之间的距离限制和关系指示词的位置限制获取候选关系三元组;然后采用全局排序和类型排序的方法来挖掘关系指示词;最后使用关系指示词和句式规则对关系三元组进行过滤.在获取大量关系三元组的同时,还保证了80%以上的微观平均准确率.

Entity relation extraction is an important task in information extraction which helps people find knowledge quickly and accurately in various text. Traditionally, entity relation extraction methods require a pre-defined set of relation types and a corpus with manual tags. But it is difficult to build a well-defined architecture of the relation types and it takes a lot of time to label a corpus. Open entity relation extraction is the task of extracting relation triples from natural language text without pre-defined relation types. There is a lot of research in the field of English open entity relation extraction, but rarely in the field of Chinese open entity relation extraction. This paper presents the UnCORE （unsupervised Chinese open entity relation extraction method for the Web）. UnCORE is an unsupervised open entity relation extraction method which discovers relation triples from large-scale Web text. UnCORE exploits using word distance and entity distance constraints to generate candidate relation triples from the raw corpus, and then adopts global ranking and domain ranking methods to discover relation words from the candidate relation triples. Finally UnCORE filters candidate relation triples by using the extracted relation words and some sentence rules. Results show that UnCORE extracts large scale relation triples at precision higher than 80%.