摘要：

针对实体间关系的表现形式和中文句法结构间存在的映射关系，提出一种新的实体关系挖掘算法。以实体关系在中文句法结构中的表现形式为基础，通过训练得到关系挖掘模板(单通道挖掘机)，将每一个待挖掘的句子交给挖掘机，利用其中包含的挖掘模板对句子进行关系挖掘。为找到更准确的关系元组，对得到的元组进行准确化处理。该算法仅根据句子的句法结构即可进行实体关系挖掘，关系挖掘类型和句法结构映射明确，拓展性较强。实验结果证明，利用中文句法结构和实体关系间的映射关系能进行关系挖掘，挖掘准确率可达到80%以上。

关键词: 关系挖掘, 单通道挖掘机, 汉语句法结构, 动态关系, 映射矩阵, 挖掘模板

Abstract:

Typically, mappings exist between entity relations and Chinese syntactic structures of sentences. A novel method Relation Extraction Based on Syntactic Structure(REBSS) is proposed to extract entity relations based on mappings between entity relations and Chinese syntactic structures. Firstly, extraction patterns are acquired according to mappings and used to construct a single-pass extractor. Then, each candidate sentence is referred to the single-pass extractor to extract entity relation. Finally, in order to achieve more accurate relation tuples, the original tuples are refined. The proposed method simply based on explicit mappings between Chinese syntactic structures and entity relations and its characteristic of scalability. Experimental results show that the relation extraction using mappings between Chinese syntactic structures and entity relations is feasible, efficient, and it gets an average precision above 80%.

Key words: relation extraction, single-pass extractor, Chinese syntactic structure, dynamic relation, mapping matrix, extraction template

中图分类号: 

• TP18