Navigated random walks on amazon book recommendation network

Abstract

Recommendation networks, where online objects such as books, songs, movies are connected to similar others via hyperlinks, have been considered of great significance for the system to distribute the web traffic, and for users to explore relevant information. However, where are the recommendation networks leading users to, and how efficient are they navigating users are still open questions. In this paper, we study the topology of, and user behaviours on a book recommendation network collected from Amazon based on a width-first search. A self-avoiding random walk model is applied to describe users’ surfing behaviour on such network. We show that the recommendation network tends to rapidly navigate users to very popular books, leading to the monopoly of traffic by these blockbusters. The niche books are therefore hidden in the dominance of the popular ones. As a consequence, the book recommendation network is found unable to quickly and accurately navigate users to their potential interests, especially these niche ones.