Empirical Analysis of the Online Rating Systems

TL;DR

This paper empirically analyzes online rating bipartite networks from Amazon and Stack Overflow, revealing growth patterns, degree distributions, and user behavior trends to understand their evolution and practical implications.

Contribution

It provides a detailed empirical study of bipartite network evolution, highlighting specific degree distribution patterns and user behavior dynamics in online rating systems.

Findings

User degree distribution follows a power law with exponential cutoff.

Large-degree objects attract more ratings than expected.

Users with higher degrees have more focused behaviors.

Abstract

This paper is to analyze the properties of evolving bipartite networks from four aspects, the growth of networks, the degree distribution, the popularity of objects and the diversity of user behaviours, leading a deep understanding on the empirical data. By empirical studies of data from the online bookstore Amazon and a question and answer site Stack Overflow, which are both rating bipartite networks, we could reveal the rules for the evolution of bipartite networks. These rules have significant meanings in practice for maintaining the operation of real systems and preparing for their future development. We find that the degree distribution of users follows a power law with an exponential cutoff. Also, according to the evolution of popularity for objects, we find that the large-degree objects tend to receive more new ratings than expected depending on their current degrees while the small-degree objects receive less ratings in terms of their degrees. Moreover, the user behaviours show such a trend that the larger degree the users have, the stronger purposes are with their behaviours except the initial periods when users choose a diversity of products to learn about what they want. Finally, we conclude with a discussion on how the bipartite network evolves, which provides guideline for meeting challenges brought by the growth of network.