In recent years, online reviews play a vital role for promoting any kind of product or services. Businesses may embed fake reviews in order to attract customers to purchase their products. They may even highlight the benefits of their own product or criticize the competition's product. Marketers, advertisers, and other online business users have incentive to create fake positive reviews for products which they want to promote or give fake negative reviews for products which they really don't like. So now-a-days writing a deceptive review is inevitable thing for promoting their own business or degrading competitor's reputation. Thus, identifying deceptive reviews is an intense and on-going research area. This research paper proposes machine learning model approach to identify deceptive reviews. The paper investigates the performance of the several experiments done on a Deceptive Opinion Spam Corpus dataset of restaurants reviews. We developed a n-gram model and max features to identify deceptive contents with a particular focus on fake reviews. Further, we conduct a benchmark study to investigate the performance of two different features extraction techniques and apply five machine learning classification techniques. The experimental results show that passive aggressive classifier outperforms other algorithms, and it reaches the highest accuracy not only in text classification but also to fake reviews. We also study the data augmentation and implement different deep learning techniques.

翻译：近年来，在线评论在推广各类产品或服务中发挥着至关重要的作用。商家可能会植入虚假评论以吸引顾客购买其产品，甚至可能夸大自家产品优势或贬低竞品。营销人员、广告商及其他在线商业用户有动机为其推广的产品编造虚假好评，或为其不认可的产品伪造恶意差评。因此，当前通过撰写欺骗性评论来推广自身业务或诋毁竞争对手声誉已成为常见手段。识别欺骗性评论已成为一个紧迫且持续的研究领域。本研究提出基于机器学习模型的欺骗性评论识别方法。论文基于餐厅评论的Deceptive Opinion Spam Corpus数据集开展多项实验，通过构建n-gram模型与最大特征量方法重点识别虚假评论中的欺骗性内容。在此基础上，我们开展基准研究，比较两种特征提取技术的性能差异，并应用五种机器学习分类算法进行性能评估。实验结果表明，被动攻击分类器在文本分类与虚假评论识别中均表现最优，达到最高准确率。同时，我们还研究了数据增强技术并实现了多种深度学习方法。