Anime Recommendation System - Collaborative Filtering Project

For this project I chose to do another anime related project. Specifically to make a recommendation system to find the similarity between shows and users to suggest comparable anime shows, users and to predicted a user's ratings.

The Data:

I used a csv file I cleaned from a previous project, where I had dropped all the hentai anime. I'm not interested in that for this project. Maybe later... ended with 5,561 unique anime data points and a few entries that were not usable. I also used another csv file from kaggle which contained User Ratings for animes, User ID's and anime ID's, this was used for the in conjunction, collaborative, with the anime dataset.

What Did I Do?

• Load and Prepared Data

  • Imported .cvs files, into pandas dataframes

  • Separated dataset to only include Anime TV shows, Didn't want to include movies.

  • Merged both dataframes on the shared animeID column. As well as drop all non-rated user ratings.

• Did some Exploratory Data Analysis:

  

  • Plotted the relationship between User Ratings and total amount of User for each rating (1-10), across all animes.

  • Visualized the Count of Ratings Per Anime (Filtered up to 300 Ratings per Anime

  • Visualized the Count of Ratings Per User (Filtered up to 300 Ratings per User

• Filtered the dataframe

  • Limited dataframe size to include value counts greater then a min value count(50) of anime_id and user_rating, mainly for computing reasons.

• Created Matrix - The start of the recommendation system

  • Pivoted the table to show User_ID's on one axis and the Anime TV Shows across the other. This allowed me to chart the User Ratings by User_ID's and Anime.

  • Transposed and converted the matrix to a sparse matrices

• Computed the the similarities

  • I ran the spared matrix and transposed matrix through cosine_similarity to find the anime and user similarities

  • Inserted the 2 similarity matrices into separate dataframes

• I made functions to call for the recommendations of:

  • Top Similar Animes
  • Top Similar Users with similarity values
  • Similar User Ratings
  • Predicted Rating of a Anime and a User specified.

• Lastly I made a Function to calculated the Mean Square Error and the Root Mean Square Error of the Predicted Rating of a Anime and a User specified, mentioned above.

Hope You'll enjoy my Anime Recommendation - Collaborative Filtering Project

Add me on LinkedIn if you liked.

Built With

• Python 3.8
• Jupyter Notebook 6.0.0
• A few of the main imports: pandas, numpy, pyplot, scipy, sklearn

Authors

• Samuel Diaz - Creator - sdman135 - LinkedIn