DieAnotherDay Casino Roya Topic 3 Topic 4 It Happened one Night Blood diamo Pulp Fictio Fahrenheit The sixth sens Hotel star Wars: Episode V- Empire sakes The Matrox 争●“ 67891011 Topic 10 The shawshank Redomatio 8910111213 1. Animation 6.Thriller cummcntar 2.ch⊥1dren 3. Fantasy Adventure 13. Musical 15. Romance Figure 1. Top 5 movies under 10 different topics and their provided genre information are similar and users and users are alike. The sub 1. There are at least 4 movies under the same genre ctive grouping result is exactly the same with that of for every topic. Topicl, Topic2 and Topic3 all have a k-nn method to the matrix Theata with the same more than two such common genres. The high co- hyper-parameters. Considering the semantic meanin occurrence frequency of different movies under the of these 24 tags, we conclude that the result of the anal- same genre reflects the large extent of conformation ysis on the user-tag matrix is persuasive. The analysis The coexistence of movie series like The matriz and on the item-tag matrix is effective in a similar way. Yet there are more merits for analyzing the item-tag ma can find not only the movies of the same genres but rix because we can obtain the names and genres of the Iso the movies of the same series movies from the overview data of the dataset 2. The five movies in Topic6 all reflect big social prob- In the second example, we extract the tag data in movie- lems. This problem could be war, terrorist attack, or lens dataset and get the e-tag matrix. There are social security crisis. This explains why these movies 600 movies in this matrix. We make the tag analysis under different genres are in the same topic. The to find the latent topics again. For the sake of leaving topic finding result is not equal to the genre classifica- more space for showing more important results, we set tion. The topic"social problem"may be interesting the desired topic number as 10. Fig. 1 presents the five for some of the users. These details are more valuable most probable movies per topic. According to it, we for inferring the users' preference than genres. have several observations as follows 3. According to the corresponding rating data, the aver- age variance of ratings of the five movies under their1.Animation 2.Children 3.Fantasy 4.Comedy 5.Crime 6.Thriller 7.Action 8.Adventure 9.Sci-Fi 10.Drama 11.Western 12.War 13.Musical 14.Mystery 15.Romance 16.Documentary 17.Horror Figure 1. Top 5 movies under 10 different topics and their provided genre information are similar and user5 and user6 are alike. The sub￾jective grouping result is exactly the same with that of a k-NN method to the matrix T heataU with the same hyper-parameters. Considering the semantic meanings of these 24 tags, we conclude that the result of the anal￾ysis on the user-tag matrix is persuasive. The analysis on the item-tag matrix is effective in a similar way. Yet, there are more merits for analyzing the item-tag ma￾trix because we can obtain the names and genres of the movies from the overview data of the dataset. In the second example, we extract the tag data in Movie￾lens dataset and get the movie-tag matrix. There are 7600 movies in this matrix. We make the tag analysis to find the latent topics again. For the sake of leaving more space for showing more important results, we set the desired topic number as 10. Fig.1 presents the five most probable movies per topic. According to it, we have several observations as follows: 1. There are at least 4 movies under the same genre for every topic. T opic1, T opic2 and T opic3 all have more than two such common genres. The high co￾occurrence frequency of different movies under the same genre reflects the large extent of conformation. The coexistence of movie series like The Matrix and Star Wars under T opic7 illustrates our tag analysis can find not only the movies of the same genres but also the movies of the same series. 2. The five movies in T opic6 all reflect big social prob￾lems. This problem could be war, terrorist attack, or social security crisis. This explains why these movies under different genres are in the same topic. The topic finding result is not equal to the genre classifica￾tion. The topic ”social problem” may be interesting for some of the users. These details are more valuable for inferring the users’ preference than genres. 3. According to the corresponding rating data, the aver￾age variance of ratings of the five movies under their