Due to increasing vast use of internet, more data is generated every day. Maintaining compatibility with this trend, companies and organizations consider the comments from their users as feedback to improve their business. Apart from this, most of the users and customers use others’ points of view to make decision about a product or an issue. Thus, more attentions are paid to opinion mining nowadays. Since internet users are easily able to comment and share their experiences on specific topics, it is likely to increase fake review attacks, called spam, through individual or groups. Hence, reviews may contain deceptive comments, so blind trust in them leads to unfavorable consequents.

Discerning a spam review from non-spam ones for human is difficult and inaccurate, therefore more attention is given to detect review spam spontaneously. Although the issue has attracted more attention so far, but it’s still an open problem and one-third of opinion reviews on web are approximated as spam. To avoid deceptive reviews effects on decision making, opinion spam detection techniques benefit machine learning methods to filter out these harmful reviews. From the primitive work on detecting deceptive reviews to now, most researches apply machine learning classifiers to distinct spam reviews from others. Consequently, more attentions are paid to learn appropriate features which lead the classifiers to have better performance.

Spam reviews are divided in to 1) untruthful reviews which deliberately affect user decisions, 2) reviews which only advertise on the brands and 3) non-reviews which are advertisements or irrelevant. In this seminar, we focus on reviews type 1, which try to mislead users by deceptive comments on a special issue. We review the existing methods and discuss about the models we’ve implemented so far.