Please use the Movielens dataset (you can find it here, see notes below about the data) and use the data to:

1. Create a collaborative filter recommender system
2. After implementing step 1, you will have a train and test dataset. Build a graph for each of the train and test datasets in which users, movies and their genres are connected. In this graph, each user, each movie and each genre has a node representing them. If user A has a rating for movie B then there is an edge in the graph connecting them. Also, if movie B has two genres: Action and Drama, then node B will be connected to nodes Action and Drama. Edges in this graph have no edge weights and no edge directions (this is an unweighted, undirected graph)
   1. What is the movie with the highest degree of centrality in train and in test datasets separately? (you choose which centrality metric is best in this case)
   2. What is the genre with the highest degree of centrality in train and in test datasets separately? (you choose which centrality metric is best in this case)
   3. use your recommender system to predict the ratings for the test dataset. Build a graph for the predictions and answer question 2.2 and 2.1 for this dataset as well

   PS: Build only Collaborative Recommender System

3. NOTES below:
4. When you go to the dataset link, there are 2 datasets, one small (named ml-latest-small.zip) and one a larger, more realistic dataset (named ml-latest.zip). The former is small and your model will suffer from lack of data, don’t use it. The latter is large and so may push your compute power depending on the machine you’re using. The best approach is to use the larger dataset and downsample it. How you do the downsampling, the ratio and the methodology is up to you (you will be asked about your choices during the presentation)
5. Feel free to give insights in comments in Jupyter Notebook

YOU DONT HAVE TO DO THE ENTIRE CODE. ONLY WORK ON QUESTIONS 2.1 and I will provide the dataframe and training and testing of it as well.