基于规则和分类模型的核心目标词识别研究

Core Target Words Identification Based on Rules and Classification

核心目标词识别是对一条待测句子,识别出句子中能够激起核心语义场景的目标词.文章把核心目标词识别任务分成基于规则过滤识别阶段和基于分类模型识别阶段.利用预先构建的词元表对待测句子中的词进行筛选,识别出候选目标词,使用基于分类模型的识别方法,构建分类特征模板,最终确定句子的核心目标词.文章在汉语框架网的标注语料集上进行测试,实验结果表明,相比于基于规则过滤识别阶段,基于分类模型识别阶段识别率有显著地提升.

For a testing sentence, core target word identification is identified the target word which can evoke core semantics scene. It is divided core target word identification into rules filter and classification. It uses pre-built lemmas table to sift the words in a sentence, identifies the candidate target words, combines the classification model and feature templates, and finally determines the core target word. Testing on Chinese frame network, the results show that the classification has been increased significantly against rules filter.