Quantifying users' selection behavior in online commercial systems

In order to uncover the online user behavior patterns, this study uses massive data from online movie rental websites as an example to explore users' behavior characteristics of watching movies and puts forward a search model to fit users' viewing mode. We use complex network tools to construct and analyze the movie space. Three main conclusions are drawn. First, the average similarity between two movies a user consecutively watched is high if this user has low activity. Second, movie stickiness increases as movie popularity increases. Third, two consecutively watched movies will not be similar if these two movies are viewed at relatively long time interval. Comparing the movie space with the product space studied by Hidalgo et al. in 2007, we find that similarity is the most important factor in both networks, but jumping behaviors which do not apply to the product space exist in the movie space. Based on the above analysis, we propose a model to simulate users' behaviors of watching movies and obtain the model parameters that best fit the real data. This model reveals users' viewing mode hidden in the data. The search model may help movie websites to recommend movies for users precisely and bring commercial benefits. It is also of great significance in film promotion and development.