Social network-based user profile data inference : a systematic method for assessing user profile reliability

Abstract

Correct user profile data are essential for successful implementation of corporate customer relationship management (CRM) and recommender systems. A company often receives incorrect user profiles or collects incomplete user information when users are unwilling to provide personal information. To address this problem, I propose a user profile quality management system. Specifically, I design and implement a cooperative query answering system that provides correct inferences about uncertain user profile information by drawing on the profile information of a user’s social network neighbors. The proposed system is based on a statistical inference model to predict the real age of users. Specifically, the system improves inference accuracy by using a top-k selection mechanism based on an entropy measure. This empirical study uses a large dataset of real user profile information and communication data to construct a social network and a user profile inference model. Through several experiments, I show that the proposed system outperforms competing models in terms of predictive power for user actual age. The experiment results has described with the concept of homophily. In addition, this paper proposes a user profile inference theory based on simulation. The goal of simulation is supporting the known limitation of real data analysis due to the existence of distorted information and checking the robustness of the proposed method. Managerial implications are discussed with respect to how a company’s legacy systems such as CRM and recommender systems can benefit from the system, specifically by maintaining correct and reliable user profile data.