结合词语规则和 SVM 模型的军事命名实体关系抽取方法?

抽取作战文书中的军事命名实体关系，是实现作战文书语义理解的一种有效方法。在分析作战文书中军事命名实体词语规则的基础上，提出了一种结合词语规则和 SVM 模型的军事命名实体关系抽取方法。首先，使用词语规则整合作战文书中连续出现的军事命名实体并抽取其关系，使其更加适合 SVM 模型。然后，使用 SVM 模型对传统规则模板难以使用的词窗、词性和距离等特征进行建模，抽取军事命名实体关系。实验结果表明，优先利用词语规则能充分提高 SVM 模型抽取军事命名实体关系的效果，与单纯使用 SVM 模型相比，准确率和召回率分别提高了8．73％和41．71％。

A Military Named Entities Relation Extraction Method Based on SVM Integrated with Word Rules

The semantic extraction of operation document can benefit the option of operation processing, and a core technol-ogy in the semantic extraction is Military Named Entities (MNEs) and their relationships discovery. In this paper, we pro-posed a MNEs relation extraction method, which integrated the word rules and SVM model. We first combined the successive MNEs with word rules, which is an important preprocessing of SVM model. Then we modeled some features of MNEs rela-tion, such as word windows, POS and distances which cannot be implemented conveniently in traditional rule template. Ex-periments show that, the performance of our method to MNEs relation extraction can be improved obviously, the precise and recall rate is 8. 73% and 41. 71% higher than simple SVM model.