Implementing a Domain-Independent Framework to Detect Suspicious Review Patterns

Nowadays detecting opinion spam has attracted a lot of attention. Different approaches have been taken to tackle this problem. However, most of these approaches are neither domain-independent nor scalable. In this paper, we focus on the similar but more general problem on how to detect abnormal patterns in reviews. These abnormal patterns can indicate potential spam activities. We will implement a scalable, domain-independent framework with the help of the Hadoop ecosystem tools. We will extract certain relationships between different fields in our data and the rating that a review gave a product or service, and in turn determine how “abnormal” these relationships are based on statistics acquired from the entire dataset. We applied this technique to three datasets in different domains and found such abnormal patterns, which indicate potential spammers and/or spam activities.