Sensei-Select

Sensei-Select is an anime recommendation service based on similarity to existing media.
Demo website here: https://seungho715.github.io/Sensei-Select/

Major Questions

• How do we create an effective Media recommendation service for a specific medium?
• How do we curate data from a database into usable features?
• Which particular features do we choose to train on, and how do we choose these features?
• How do we balance model complexity with accurate recommendations?

Key Findings And Considerations

• The model recommends anime based on cosine similarity of the features, but can recommend inappropriate titles when searching for safe anime similarity.
• We also found that we encode features using tf-idf encoding for textual features and one hot encoding for category features.
• Popular anime tend to recommend other popular anime. How can we incorporate more niche recommendations?
• An anime often recommends variations of itself. For instance, Hunter x Hunter (2011) recommends the 1999 release. How can we form a model to only recommend similar but distinct releases?
• Language-specific queries often result in language-specific output. For instance, The Nine Heavens Swordmaster, a Korean release, results in many Korean, Japanese, and Chinese animes.
• It's crucial to filter out features to avoid overfitting. Using too many features in model training can lead to unexpected and unwanted recommendation results.

Instructions for setting up Backend environment and running

1. Create a Python virtual Environment and install the following packages with pip: numpy pandas tensorflow matplotlib sklearn flask
2. Run flask --app Backend run
3. Now, you can make a GET request to the http backend with cURL: curl -G -d "title=Naruto" http://localhost:5000/get_anime_recommendation/. Replace the title parameter with the Romaji name of an anime.

Instructions for setting up Frontend environment and running

1. Either remove or change the "homepage": "https://seungho715.github.io/Sensei-Select" field in the anime_frontend/package.json file on line 4 to "homepage": "http://localhost:3000"
2. Within the anime_frontend directory, run the command npm install then npm run start
3. Log onto localhost:3000 and type in the name of an Anime
4. Click the Search button to select the particular title of your Anime
5. Click the Recommend Me button and scroll to the bottom of the page (after an approximately 10 second loading time) to view your top 10 recommendations
6. Click on the title of the Anime to go to its Anilist page

Data Set Used

Our data set was gathered from Anilist.co using their GraphQL API. This includes extensive information relating to each anime, including tags, characters, voice actors, staff, studios, etc. All of the data scraped has been formatted into tables that can be found in the Tables folder.

The majority of our data has been collect throughout April 2024. As Anilist will continue to update their database, the tables saved here will eventually become outdated.