With the prevalence of e-commerce, online product reviews are increasingly considered crowd-sourced consumer opinions that significantly influence customer purchasing decisions and product rankings. It is therefore important to ensure the truthfulness of reviews by detecting and filtering out fake/spam reviews. This article presents an effective framework to analyze review credibility for spam detection and opinion mining. It incorporates three methods: duplicated review detection, anomaly detection, and incentivized review detection, that complement each other to produce statistical credibility scores indicating review credibility. A practical end-to-end system is designed and developed accordingly, and is equipped with high-level data visualization for easy interpretation and summarization of the analysis results. Experiments on an Amazon review dataset demonstrate its efficiency, scalability and accuracy. This system could help e-commerce and consumers identify fake reviews, refine product rankings, and constrain vendors and spammers from engaging in dishonest practices.
