基于CBC-LIKE算法的产品特征词聚类的研究

摘要

Aiming at the problem of the various product feature expressions existing in user reviews,it's necessary to cluster the product feature words in the task of fine-grained opinion mining.According to the calculation characteristics of different semantic similarities,a semantic similarity mixture calculation method based on semantic knowledge and context entropy model is proposed to calculate the extract the semantic similarity of feature words.The traditional CBC algorithm is improved.A CBCLIKE method suitable for product feature words clustering is proposed.The experiment is conducted for the real review corpus in three domains.The performances of the proposed semantic similarity calculation method and clustering algorithm are analyzed.The experimental results show that the method is effective,its performance is better than that of other two benchmark methods,which has perfect effect.%用户评论中存在产品特征表达多样性问题,在细粒度观点挖掘任务中需要对产品特征词聚类.首先,结合不同的语义相似度计算的特点,提出基于语义知识和上下文熵模型的语义相似度混合计算方法,计算抽取得到的特征词语义相似度;然后改进了传统CBC算法,提出适用于产品特征词聚类的CBC-LIKE方法实现聚类.最后在三个领域的真实评论语料上进行实验,对提出的语义相似度计算方法和聚类算法的性能进行了分析.实验结果表明,所提方法是有效的,与另外两种基线方法相比性能较优,取得了较好效果.