基于信息增益的Web人物关系抽取

Information gain-based method of people relationship extraction on Web

针对人物关系抽取中的效率与准确性问题进行了研究,提出一种基于信息增益的轻量级Web人物社会关系提取方法。它通过计算初始关系元组的关系描述词的信息增益值进而确定元组上下文位置并据此创建相应的关系抽取模板,最后利用模板实现了Web的人物关系自动提取。针对中文语义上存在相似性的问题,引入了基于《同义词词林》与基于知网的人物关系描述词扩展方法。对于某一句子内包含多个人物实体且存在多种人物关系的情况,提出了一种基于模板上下文信息增益值模糊匹配的方法来抽取符合特定人物关系的人物实体。实验结果证明该方法的平均准确率为89.92%,平均召回率为84.64%。基于信息增益的Web社交网络人物关系抽取方法能有效地完成实时语料中的关系抽取任务。

For the problem of accuracy and efficiency in the people relationship extraction,this paper presented a lightweight Web people＇ s social relations extraction method based on the information gain. It calculated the information gain of relationship description word in the initial relationship tuple and then located the tuple context. Moreover, it created a corresponding tem- plate for the Web automatic relationship extraction. In view of the circumstance of the Chinese semantic similarities, this paper introduced the method that people relationship description words extension based to the HowNet and ＂Chinese Thesaurus＂. Sometimes one sentence contains more than two entities and various kinds of relation. This paper presented a new approach to this situation which extracted people entities that met certain condition by matching template context information gain value faintly. Test result shows that this method＇ s average accuracy rate is 89.92% , the average recall rate is 84.64%. The method of people relationship in Web social network extraction based on information gain can accomplish the task of people relationship extraction in the real-time data effectively.