Affiliation:
1. Data Mining and Web search Research Group, Informatics Laboratory; Computer and Automation Research Institute of the Hungarian Academy of Sciences
Abstract
KDD Cup 2007 focuses on predicting aspects of movie rating behavior. We present our prediction method for Task 1 "Who Rated What in 2006" where the goal is to predict which users rated which movies in 2006. We use the combination of the following methods, listed in the order of their accuracy:
• The predicted number of ratings for each movie based on time series analysis, also using movie and DVD release dates and movie series detection by the edit distance of the titles.
• The predicted number of ratings by each user by using the fact that ratings were sampled proportional to the margin.
• The low rank approximation of the 0-1 matrix of known user-movie pairs with rating.
• Prediction by using the movie-movie similarity matrix.
• Association rules obtained by frequent sequence mining of user ratings considered as ordered itemsets.
By combining the predictions by linear regression we obtained a prediction with root mean squared error 0.256. The first runner up result was 0.263 while a pure all zeroes prediction already gives 0.279, indicating the hardness of the task.
Publisher
Association for Computing Machinery (ACM)
Reference11 articles.
1. New light on the correlation coefficient and its transforms;Hotelling H.;Journal of the Royal Statistical Society B,1953
Cited by
1 articles.
订阅此论文施引文献
订阅此论文施引文献,注册后可以免费订阅5篇论文的施引文献,订阅后可以查看论文全部施引文献