MLP Live session

This live session, part of the Machine Learning Practice series, focuses on recommendation systems, specifically collaborative filtering techniques. The instructor begins by introducing the concept of collaborative filtering and its three main types: item-item, user-user, and matrix factorization. He explains that item-item filtering is more stable than user-user because user preferences change over time, while items remain finite. The session then moves to a practical demonstration using the MovieLens dataset. The instructor shows how to create a user-item matrix, handle missing values by filling with zeros, and compute cosine similarity between items to find similar movies. He illustrates this with an example, recommending movies similar to ‘Charlie and the Chocolate Factory’. The discussion then shifts to user-user collaborative filtering, where the data is structured with users as rows and movies as columns. The instructor explains the prediction formula, which involves weighting ratings by similarity scores and normalizing. He also addresses a question about the difference between item-item and user-user approaches. Finally, the session introduces matrix factorization, a more advanced learning-based approach that maps users and items to a lower-dimensional latent factor space. The instructor explains the concept of randomly initializing matrices and multiplying them to predict ratings. However, the session is interrupted by unrelated questions and technical issues, and the instructor admits to incomplete notebooks and errors. The session ends with a promise to share updated notebooks.