A mathematical model to cluster reviewers for online review system

Abstract

Online business models accept reviews or feedback from customers which are processed and analyzed for important business decisions. Online reviews are helpful to understand the usefulness or popularity of a product. However, it has been observed that sometimes fake reviews are frequently used to boost the popularity of one's own product or to damage reputation of competitors' products. Henceforth it is an interesting research problem to validate reviews or trustworthiness of reviewers. In this paper, a mathematical model is introduced to rate and cluster reviewers based on relevant parameters. It has been observed from business intelligence perspective, that grouping reviewers into different clusters, rather than ranking them individually based on their authenticity, would be more beneficial for potential buyers to understand the quality of reviewers. In the proposed model, clustering is performed using two weighted scores based on average opinion variance and product price. The mean shift clustering algorithm is used to dynamically slab the product price attribute while Jenks Natural Breaks Optimization (JNBO) method and K-means algorithm are applied for the reviewer clustering. Further this research work analyses the impact of product price on reviewer rating and validates the result using t-test statistical method. The proposed methodology is experimented on Amazon datasets to show efficacy of the model.