BERT Anime Recommender

BERT Anime Recommender

BERT Anime Recommender is a unique recommendation system designed to help anime fans discover new shows they might enjoy. Unlike traditional systems that rely on user ratings, this tool uses a BERT transformer model to analyze users' favorite anime titles. By focusing on what users genuinely like, it aims to provide more personalized and accurate recommendations.

Benefits

BERT Anime Recommender offers several key advantages:

• Personalized Recommendations: The system generates tailored suggestions based on users' favorite anime, ensuring a more accurate match to individual tastes.
• Cold Start Problem Solution: It effectively handles the cold start problem by providing meaningful recommendations even with limited user data.
• Entertainment Focus: Developed purely for entertainment, it helps users discover new anime and enhance their viewing experience.
• Deep Learning Technology: Utilizes advanced NLP models like BERT to capture deeper patterns of preference.

Use Cases

This recommender system is ideal for anime enthusiasts looking to expand their watchlist. Whether users are new to anime or seasoned fans, the tool can suggest shows that align with their preferences. It is particularly useful for those who find generic recommendations unhelpful and seek a more personalized approach.

Vibes

The creator of BERT Anime Recommender tested the system with their own favorite anime and was impressed with the results. Out of the top 15 recommendations, six were anime they had already watched and loved. The remaining suggestions, such asFrieren,Tonikaku Kawaii,Solo Leveling, andHeavenly Delusion, fit their taste almost perfectly. This demonstrates the system's ability to provide accurate and relevant recommendations.

Additional Information

The BERT Anime Recommender project was built using a dataset of over 4 million user anime lists scraped from AniList, Kitsu, and MyAnimeList. After filtering for users with at least 10 ratings, the dataset was narrowed down to around 1.5 million users. For faster training and prototyping, a subset of 600,000 users was used. The model was trained on a dataset with 54 million ratings, ensuring high-quality recommendations.

The project is open-source, and users can try the model for themselves. Setup and inference instructions are available on the GitHub repository: https://github.com/MRamazan/AnimeRecBERT.