A Multidimensional Trust Evaluation Framework for Online Social Networks Based on Machine Learning

Abstract

Due to the openness of online social networks (OSNs), they have become the most popular platforms for people to communicate with others in the expectation of sharing their opinions in a trustworthy environment. However, individuals are often exposed to a wide range of risks posed by malicious users who spread various fake information to achieve their vicious goals, which makes the concept of trust a vital issue. Most of the existing research attempts to construct a trust network among users, whereas only a few studies pay attention to analyzing their features. In this paper, we propose a trust evaluation framework based on machine learning to facilitate human decision making by extensively considering multiple trust-related user features and criteria. We first divide user features into four groups according to the empirical analysis, including profile-based features, behavior-based features, feedback-based features, and link-based features. Then, we design a lightweight feature selection approach to evaluate the effectiveness of every single feature and find out the optimal combination of features from users' online records. We formalize trust analysis as a classification problem to simplify the verification process. We compare the performance of our features with four other feature sets proposed in the existing research. Moreover, four traditional trust evaluation methods are employed to compare with our machine learning based methods. Experiments conducted on a real-world dataset show that the overall performance of our features and methods is superior to the other existing features and traditional approaches.

Keywords

• Trust evaluation
• multidimensional features
• feature selection
• machine learning
• social networks