Abstract: With the development of electronic commerce, assessing the trustworthiness of reviews is becoming a key issue. Heuristic strategies or traditional supervised learning methods cannot effectively solve this task. There must be some differences on language structure and sentiment polarity between deceptive reviews and truthful ones. This paper defines the features related to the review text and uses genetic algorithm for the features selection of language structure and sentiment polarity. Then, this paper uses the selected features and combines two non-supervision clustering algorithms to identify deceptive reviews. The experimental results verify the effectiveness of the proposed methods.

Key words: deceptive reviews, clustering, language structure, sentiment polarity, genetic algorithm

摘要： 随着电子商务的发展，识别网络中的虚假评论意义重大。传统的启发式策略或全监督学习算法不能有效地解决该问题。虚假评论与真实评论在语言结构和情感极性上存在差异，提出基于遗传算法对语言结构及情感极性特征进行优化选择，并利用选取的特征结合无监督硬、软聚类算法对虚假评论进行识别。实验结果验证了所提算法的有效性。

关键词: 虚假评论, 聚类, 语言结构, 情感极性, 遗传算法