基于多分类器的金融领域多元关系信息抽取算法

Multi-relation extraction in finance based on multi-classifier

为深入分析金融领域文本信息给投资决策提供支持,研究了从中文文本中识别收购类事件描述句及抽取事件角色(即识别关系及关系的元)相关问题。在事件句的识别上,提出了基于SVM的有监督算法。对于关系识别及关系元的抽取,针对多元关系的特点,分别设计了单分类器的算法和多分类器的算法,单分类器的算法由一个分类器负责识别多元关系的所有角色,而多分类器算法使用不同的分类器来识别具有不同语义约束的角色。实验结果表明,多分类器的算法明显优于单分类的算法,角色识别的F-Measure可以提高1.9%。

To analyze financial text information and support investment decision, event description sentence detection of purchasing and selling assets and event role identification （recognizing relation and relation elements） from Chinese texts are studied. An algorithm based on SVM is investigated to detect event description sentences. Considering the characteristics of multi-relation, two algorithms are designed to identify relation and reiation elements. One is based on a single-classifier which identifies all the roles of multi-relation through one classifier, and the other is based on a multi-classifier which designs a classifier for each semantic role separately. The experimental results demonstrate that the multi-classifier algorithm performs better than the mono-classifier algorithm, and the F-Measure of event role identification can be improved by 1.9%.