In order to evaluate the prediction accuracy of our approach, we used the EachMovie dataset of the Digital Equipment Corporation for experiments. The EachMovie dataset consists of 2,811,983 preferences for 1,628 movies rated by 72,916 users explicitly. The user preferences are represented by means of numeric values from 0 to 1.0 at 0.2 intervals, i.e., 0.0, 0.2, 0.4, 0.6, 0.8, 1.0. In the EachMovie dataset, the attributes of an item are the genre of a movie. There are 10 different genres such as action, animation, art-foreign, classic, comedy, drama, family, horror, romance, and thriller.
For our experiment, we retrieved 3,763 users who rated at least 100 movies among all the users in the dataset. We have chosen randomly 10 users out of 3,763 users as the test users and the rest 3,753 users as the training users. For each test user, we chose 5 movies randomly that are actually rated by the test user as the test movies. The final experimental results are averaged over the results of 5 different test sets. So the total number of experiments is 250.