基于多模板模糊竞争的涉案财物关系抽取方法

Relation extraction method for property involved based on multiple templates and fuzzy competition

摘要:

从法律法规中自动抽取涉案财物处置关系并构建知识库，对司法实践中涉案财物的智能管理有着重要意义.但该场景下训练语料少且单一、抽取准确率要求高，机器学习的关系抽取方法不适用.针对这个问题，基于模板匹配的关系抽取思想，结合模糊逻辑机制，研究了一种适用于涉案财物关系抽取的方法.该方法考虑了涉案财物知识库的实际需求，在传统三元组关系的基础上增加了二元属性，扩充为五元组关系模式，从不同维度设计了三个抽取模板，并借助模糊逻辑对抽取的结果进行评判，竞争出效果较优的五元组关系，为涉案财物知识库的构建提供了支持.

Abstract:

Extracting the relation of the property involved from laws and regulations to build a knowledge base is of great significance to the intelligent management of the property involved in the judicial practice. However, in this scenario, the training corpus is small and single, while high extraction accuracy required, so the relation extraction method of machine learning is not applicable. Aiming at addressing this problem, we propose a new relation extraction approach that extracts relations by template matching with fuzzy logic mechanism. In this approach, in order to cater to the actual needs of the knowledge base for property involved, two attributes are added to the traditional threetuple relation to obtain the five-tuple relation model. Three extraction templates were designed from different dimensions, and the extraction results were judged with the help of fuzzy logic to compete for a better five-tuple relation, which provided support for the construction of the knowledge base for property involved.