Declarative Programming Approach for Fake Review Detection

Online reviews play an essential role in our daily life. Thus, approaches for detecting fake reviews are of high demand. This paper presents an approach to detect fake reviews incorporating the behavior of authors of reviews combined with properties derived from the content of their reviews. We aim to design a white-box approach which is becoming a major requirement nowadays in the industry. This is due to the fact that there are increasing social concerns about decisions made based on personal information. In other words, we seek to design a white-box model that can let users understand what is going on regarding their personal data. In contrast to blackbox models, such as deep-learning that are hard to be explained in general. Consequently, we propose a rule-based fake review detection system using Answer Set Programming (ASP) which is a powerful tool to declare malicious behavior patterns specified via a variety of constraints. This way we can create powerful models that combine, e.g., information about the number of reviews, the number of dislikes, the analysis of the points in time reviews have been written, qualitative properties of the content based on similarity measures and derived classification of reviews and products. Such models encode the problem phrased "which reviews are to be considered genuine, fake, or need to be investigated further on" and can be used to compute an optimal solution by applying ASP techniques.