Identifying Suspicious Users and Products to Predict New Opinions

An opinion network describes how a person values an object, which can be another person or product. This type of network is ubiquitous because humans constantly evaluate their surroundings. Analyzing an opinion network can bring deeper understanding of what and how people are thinking, but at the same time, a malicious attempt to influence such network can be harmful to the public. One problem naturally arising while studying opinion networks is identifying users leaving intentionally fake opinions to influence the opinions of others. Another question that can be asked is how to predict a person’s opinion towards another. In this work, a solution to those issues is suggested using the Rev2 algorithm (developed by S. Kumar, et al) and the principle of maximum entropy. The Rev2 algorithm is used to assign a suspiciousness score to users based on how much the user’s action deviates from the rest. Also, a quality index will be assigned to products based on reliable opinions. Using this information, the principle of maximum entropy is invoked to produce the most unbiased prediction of new opinions. The proposed solution is tested against public network data including a user-to-user trust network of Bitcoin platform users and a user-to-product rating network of Netflix users.