论文部分内容阅读
【目的】针对中文在线评论产品特征与观点抽取问题,提出一种基于置信度排序模型的抽取方法。【方法】在改进HITS算法基础上,综合考虑候选特征观点词的关联关系和语义关系构建置信度排序模型,提取并过滤特征观点词。【结果】和基准模型相比,本文方法对中文语料的产品特征和观点抽取能达到较高准确率和召回率。【局限】仅针对产品显性特征抽取,没有考虑隐性特征的识别与抽取。【结论】利用特征词和观点词的双向增强关系和语义关系,可以有效抽取产品特征观点;情感极性过滤对提升观点词抽取准确率有较大作用。
【Objective】 In view of the product features and opinion extraction of Chinese online reviews, a method based on confidence ranking model is proposed. 【Method】 On the basis of improving HITS algorithm, we constructed a confidence ranking model considering the association and semantic relations of candidate feature opinion words, and extracted and filtered feature opinion words. [Results] Compared with the benchmark model, this method can achieve higher accuracy and recall rate for Chinese corpus product features and viewpoint extraction. [Limitations] only for product dominant feature extraction, did not consider the identification and extraction of hidden features. 【Conclusion】 With the two-way enhancement relation and semantic relation between feature words and opinion words, the feature points of product features can be effectively extracted. Emotional polarity filtering plays an important role in improving the accuracy of opinion word extraction.