Nowadays, a huge amount of opinions are posted and tweeted on the Web. Such opinions are a very important source of information for customers and companies. A lot of researchers describe that users are relying on online opinions to make their purchase decisions. Unfortunately, due to the business that is behind, there is an increasing number of deceptive opinions in order to deceive consumers by promoting a low quality product (positive deceptive opinions) or criticizing a potentially better quality product (negative deceptive opinions). This thesis will focus on detection of negative deceptive opinions from a negative tweet on specific brands. We applied lexical, personal profile and personal behavioral features to detect negative deceptive opinions using supervised machine learning classifiers, i.e. Support Vector Machine (SVM), Naïve Bayes, Decision Tree, Maximum Entropy, BAGGING and Random Forest. We tested our method using user opinions about different Samsung products and related issues that are collected from five official twitter accounts. One of the challenges in evaluating such a system is the lack of a large-scale labeled dataset. To resolve this issue, we construct our own dataset by recruiting multiple people to label the collected tweets. The labels are assigned by the majority vote. The acquired results indicate that our proposed system accomplishes 100% exactness with maximum entropy and 98% utilizing Naïve Bayes on our first kinds of datasets that consist of tweets having unanimous labels by all examiners, and 94% and 91% on the full labeled dataset. It is a promising approach for detecting deceptive opinions. Our approach also can help to identify defamers by analyzing the profile information of users, comment giving behavior and writing style of each user.
