基于多层关系图模型的中文评价对象与评价词抽取方法

发布时间：2018-05-05 16:36

本文选题：倾向性分析 + 观点挖掘　；参考：《自动化学报》2017年03期

【摘要】：中文评价对象与评价词抽取是文本倾向性分析的重要问题.如何利用评价对象与评价词之间的语法、共现等关系设计模型是提高抽取精度的关键.本文提出了一种基于多层关系图模型的中文评价对象与评价词抽取方法.该方法首先利用词对齐模型抽取评价对象与评价词搭配;然后,考虑评价对象与评价词的依存句法关系、评价对象内部的共现关系和评价词内部的共现关系,建立多层情感关系图,接着利用随机游走方法计算候选评价对象与评价词的置信度;最后,选取置信度高的候选评价对象与评价词作为输出.实验结果表明,与现有的方法相比,本文所提出的方法不仅对评价对象和评价词的抽取精度均有显著提升,而且具有良好的鲁棒性.
[Abstract]:Chinese evaluation object and word extraction is an important problem in text orientation analysis. How to make use of the grammar and co-occurrence model between the evaluation object and the evaluation word is the key to improve the extraction accuracy. In this paper, a method of extracting Chinese evaluation objects and words based on multilevel relational graph model is proposed. The method firstly extracts the collocation between evaluation object and evaluation word by word alignment model, and then considers the dependent syntactic relation between evaluation object and evaluation word, the co-occurrence relation within evaluation object and the co-occurrence relationship within evaluation word. The multilevel affective relationship graph is established, and then the confidence degree of candidate evaluation object and evaluation word is calculated by random walk method. Finally, the candidate evaluation object and evaluation word with high confidence degree are selected as the output. The experimental results show that compared with the existing methods, the proposed method not only improves the accuracy of evaluation object and evaluation term extraction significantly, but also has good robustness.
【作者单位】：福州大学数学与计算机科学学院;福建省网络计算与智能信息处理重点实验室(福州大学);福建江夏学院电子信息科学学院;中国科学院计算技术研究所网络数据科学与技术重点实验室;
【基金】：国家自然科学基金青年项目(61300105) 中国科学院网络数据科学与技术重点实验室开放基金课题(CASNDST20140X)资助~~
【分类号】：TP391.1

【相似文献】