A study of reliability patterns by simultaneous clustering
Jie Wang · 2014
With the growth of users' ability to create and publish content online, online reviews have become a valuable and influential tool in the decision-making process. However, due to increased competition, delivery of trustworthy review service is challenging for a website because businesses can manipulate their own or their competitor's reviews. The review reliability issue is studied quantitatively here from the data mining perspective. A simultaneous clustering is performed for both reviewers and products by applying a matrix factorization approach. A tri-factorization algorithm is proposed to factorize a data model into three factor matrices and these sub-matrices can reveal review reliability patterns. The results also can rank reviewer groups in terms of products. This approach provides a new framework for examining the reputation of past reviews and predicting the helpfulness of reviewers for new products.