摘要
目前的实体关系识别方法,无法充分利用海量未标注军事文本中的潜在信息,并且在实体关系特征词提取方面并不完善。于是本文对实体关系描述词的特点进行了分析,提出了一种语义层面的弱监督军事实体关系识别技术。本方法采取Word2vec模型和FPGrowth算法对海量未标注军事文本进行分析,从关联分析角度进行实体关系描述词提取,从语义相似角度进行关系词扩充,并提出一种浅层关系分类矩阵对实体关系进行分类。该方法在对1000篇军事文本测试中,取得了加权平均值F-Score为89.2%的效果。
The current method of entity identification can not make full use of the potential information in the military text, and it is not perfect in the extraction of the entity relation feature words. So this article analyzes the characteristics of the entity relation descriptive word, and puts forward a semantic level of weak supervision of military entity relationship recognition technology. In this method, the Word2vec model and the FPGrowth algorithm are used to analyze the massive non-marked military texts. Entity relations are extracted from the perspective of relational analysis, and the relational words are expanded from the perspective of semantic similarity. A shallow relation classification matrix is proposed to analyze the relationship classification. The method achieved a weighted average F-Score of 89.2% for 1000 military text tests.
出处
《电子设计工程》
2018年第1期74-78,83,共6页
Electronic Design Engineering
作者简介
李煜甫（1992-）,男,陕西西安人,硕士研究生.研究方向：机器学习,数据挖掘.
