Predicting IMDB movie ratings using Google Trends

Movie ratings are influenced by many factors, so accurate prediction of new movie ratings can be challenging. In recent years, various semantic analysis techniques were successfully applied to analyzing user reviews, which in turn were applied to predict IMDB movie ratings (based on IMDB reviews, youtube movie trailer reviews etc). To the best of our knowledge there has been no research done on predicting IMDB movie ratings using Google search frequencies for movie related information. However, several authors have suggested to use search volume to predict consumer behavior [1, 2]. Our main idea in this project is to characterize each movie by the set of features, and then use Google search frequencies of these features to predict movie popularity. The intuition behind this approach is that for popular movies one should see the higher search volume of queries associated with the movie. We view this problem as a supervised learning problem, and we used logistic regression, SVM and multi layer perceptron algorithms for our project. We also used dimensionality reduction techniques to select the optimum set of features for one of the experiments that we performed. In what follows we first describe the dataset that we used for our experiments. After that we give details of the experiments that we performed and discuss their results.