Show HN: I scraped 3B Goodreads reviews to train a better recommendation model (book.sv)

Article summary

The author has created a book recommendation model by scraping 3 billion Goodreads reviews and training a machine learning model. The model can recommend books based on a user's input of books they have read. The website allows users to input multiple books and find recommendations, and also has an 'intersect' feature to find users who have read the same books. The model is trained to predict the next book read in a sequence.

Main themes

• Book recommendation systems
• Machine learning models
• Goodreads data scraping
• User-based recommendations
• Sequential recommendation
• Model training and evaluation

What commenters say

• The recommendation model works well in suggesting books that users have already read and enjoyed, but may not be as effective in discovering new books or genres.
• The model's tendency to recommend books in the same series as the input books can be limiting and may not provide diverse recommendations.
• Some users find the 'intersect' feature to be useful in finding users with similar reading interests, but others have difficulty finding matches due to the strict criteria.
• The model's performance can be improved by incorporating negative signals, filtering out spam books, and using more advanced recommendation algorithms.
• The use of Goodreads data may be biased towards certain genres or types of books, and may not be representative of all readers' preferences.
• The model's ability to recommend books based on sequential reading patterns is a key strength, but may also lead to recommendations that are too similar to the input books.
• Some users appreciate the model's ability to recommend books that are not necessarily popular or well-known, but rather tailored to their individual reading interests.
• The model's evaluation metrics and performance may be improved by using more advanced techniques, such as semantic IDs and VAEs.