https://www.amazon.com/Deep-Learning-Coders-fastai-PyTorch/dp/1492045527

San Diego State University

This methodology is frequently used in recommendation problems. The examples  include deciding which stories to show on social media feed, which movies to suggest to user and many more.  In simple terms **collaborative filtering** finds what particular user liked and tries to find users with similar preferences and based on this information it would recommend to the current user the items that these other users liked. 

Collaborative Filtering

Movielens dataset contains millions of movie ratings data, namely, these are the combination of movie and user IDs and numerical rating. This dataset can be used to practice for recommendation problems. 

MovieLens Data for Collaborative Filtering

**Bootstrapping problem** is one of the major weaknesses of collaborative filtering. It means that the model / system cannot make any recommendations or draw any inferences for users or items about which it has not yet gathered sufficient information. It’s also called the cold start problem. There are several different approaches to solve this issue.  For instance, we can come up with an average embedding for a user or movie. Another option can be to select a particular user/movie to represent the average user/movie. Additionally, we can define several questions that could help initialize the embedding vectors for new users and movies.

Bootstrapping Problem in Collaborative Filtering

In collaborative filtering models, bias terms compensate for the baseline tendencies of specific users or items. Item bias accounts for items that systematically receive exceptionally high or low ratings across the board. User bias accounts for users who consistently give higher or lower ratings than the average user, adjusting for these systematic deviations.

Bias in Collaborative Filtering Models

Latent factors are features that are not supplied directly to a collaborative-filtering model but are learned from rating data during training. Each user and item is assigned a latent-factor vector, and the number of factors is chosen before training. The vectors are initialized, used to predict ratings, and updated through optimization to reduce prediction loss.

Learn Before

Related