产品评论中的隐式属性抽取研究

Implicit Feature Identification in Product Reviews

【目的】产品领域的意见挖掘是近年来的一个非常热门的话题,意见挖掘结果可以帮助过滤有害信息、进行社会舆情分析、指导用户消费和帮助商家改善产品性能等,而隐式产品属性在网络评论句中十分常见且挖掘难度大,因此对其进行研究有重要的意义。【方法】利用仅包含显式属性的某品牌汽车评论句确定多词性精简意见词,并利用同义词词林进行扩展形成意见簇,同时基于领域常用语确定属性词,并通过搭配关系计算权重,生成记录形如"{属性,意见,权重}"的字典,利用多策略隐式属性抽取算法以字典为基础抽取隐式属性,同时考虑待匹配意见词与字典中的意见词之间的相似度。【结果】可以行之有效地抽取出评论句中的隐式属性,F值达到75.55%,属于隐式产品属性抽取现有研究的较好结果。【局限】前期数据标注工作主要靠人工,较为费时费力。【结论】实验结果表明本文算法效果较好,具有一定的实用价值。

[Objective] Opinion mining in product areas draws more and more attention and becomes a hot research topic. The outcome of opinion mining can be used widely just like harmful information filtering, society opinion analysis, user consumption guidance and product improvement and so on. Implicit feature identification plays an important role because implicit features are common in network comments and the identification of them is difficult. [Methods] This paper uses the comments against a certain automobile brand which only have the explicit features to get refined multi-POS opinions and generate opinion clusters by using Synonyms Forests. Meanwhile identify opinions based on field common phrases. Dictionary in the form of {Feature, Opinion, Weight} is generated by using features and opinions, and the weight is calculated. Then deploy explicitly multi-strategy property extraction algorithm based on a dictionary and consider similarity of the opinions in unmatched comments including implicit features and dictionary. [Results] Implicit features can be extracted effectively and the F-value is 75.55% which reaches the good result of the identification of implicit features. [Limitations] Data labeling is a time-consuming job. [Conclusions] Experiment of the new algorithm shows positive result and has some practical value.