Personal Information Prediction Based on Movie Rating Data

Abstract

Movies are a major form of entertainment in the US. There are a dozens of websites focusing on movie information. On most of the websites, ratings and reviews from the users play an important role. When a user gives a movie a certain score, the user not only reflects his taste toward that movie but also potentially exposes his personal information. In this paper, we investigated several movie genres. In each genre, movies were classified into different clusters by using expectationmaximization (EM) algorithm. The classification criteria were built upon audience movie rating scores and existing user information. As a result, a new or anonymous users personal information could be predicted when he rated movies on movie-related websites. Moreover, newly released movies could be easily classified into corresponding clusters to assistant user information discovery. The revealed personal information was very useful and could be utilized in different ways such as increasing the accuracy for delivering user-related ads.