Multi-Dimensional User Credibility Analysis On Review Content

Traditionally, user credibility evaluation is calculated by comparing the overall measurements in rating systems. Actually, a user's rating against an item is different among aspects discussed in reviews. Then user credibility shall be discussed on aspect dimension instead of the overall evaluation dimension. To address this problem, we design a two-level model to analyze user credibility on aspects. Our method first detects all the involved aspects, and explores ratings for each aspect ratings by mining semantics in reviews. It makes full use of fine-grained information contains in comment text and can sensitively capture fluctuates among aspects. We model user scoring relationship by a weighted bipartite graph with users and aspects as nodes and credibility as weights. An iteration algorithm is designed for credibility calculation on the graph. We perform extensive experiments to demonstrate the advantages of our design.