Analysis and Detection of Online Paid Restaurant Reviews

In recent years, people get used to sharing their opinions and experiences on the Internet. These opinions greatly influence our decisions. For example, most people read online reviews before they make purchases. Malicious companies or individuals make use of fake reviews to control the opinions on social media and blogs. In this thesis, we collect paid reviews on Pixnet and understand this type of promotion campaigns. Some characteristics of paid reviews and writers are found. We then propose a set of features based on our observation and detect paid reviews and writers using supervised machine learning techniques. Our results demonstrate the effectiveness of features and outperform random baseline significantly. Furthermore, a collective detection method using Markov Random Fields is proposed to detect paid reviews and writers seamlessly. The collective detection method can utilize the relations among review and user instances. The results outperform the performance of separate detections.