Explains how collaborative filtering is used in recommender systems. Collaborative filtering checks on users preferences to a certain item, such as music or movies.
One of the best/most accurate representation method to use is Centered Cosine Similarity/Pearson Correlation, which normalizes the items being compared by subtracting the row mean for what value is being compared, such as listens or ratings (preference). The cosine then can be computed by using the centered values to express similarities and differences between users and their preferences. This is user-user collaborative filtering.
Another method is to make rating predictions for a user by using a matrix
Item-item collaborative filtering is a dual approach to user-user CF. Use an item type and find other similar items, and estimate listens or ratings (preference) for said item based on similar items. The same similarity metrics and prediction functions for user-user. Function/Equation @ 15:09. It predicts the rating for item i and we find the neighborhood of items that are both rated by a same user x. Then the same weighted average formula to predict user x and item i.
In a utility matrix (16:12 - identify all movies that a user has rated. Then select a movie to predict, that movie's weighted avg. will be 1.00. Then identify all movies similar to that selected movie by using Pearson correlation then compute similarities between rows. When you find the similar weighted averages, to 1.00, use the equation, @ 19:13, to predict the rating for that original specified movie.
Item similarity is more meaningful because items generally belong to a small set. Users tend to have varied taste.
name: 'Article' about: 'Articles and other documents. ' title: '' labels: Article assignees: ''
Title
URL
Summary
Key Points
Citation
Repo link
name: 'Audio Recording' about: 'Podcasts and other audio-only resources. ' title: '' labels: Audio assignees: ''
Title
URL
Summary
Key Points
Citation
Repo link
name: 'Video Recording' about: 'Video resources. ' title: '' labels: Video assignees: ''
Title
Lecture 43 — Collaborative Filtering | Stanford University
URL
https://www.youtube.com/watch?v=h9gpufJFF-0
Summary
Explains how collaborative filtering is used in recommender systems. Collaborative filtering checks on users preferences to a certain item, such as music or movies.
One of the best/most accurate representation method to use is Centered Cosine Similarity/Pearson Correlation, which normalizes the items being compared by subtracting the row mean for what value is being compared, such as listens or ratings (preference). The cosine then can be computed by using the centered values to express similarities and differences between users and their preferences. This is user-user collaborative filtering.
Another method is to make rating predictions for a user by using a matrix
Item-item collaborative filtering is a dual approach to user-user CF. Use an item type and find other similar items, and estimate listens or ratings (preference) for said item based on similar items. The same similarity metrics and prediction functions for user-user. Function/Equation @ 15:09. It predicts the rating for item i and we find the neighborhood of items that are both rated by a same user x. Then the same weighted average formula to predict user x and item i.
In a utility matrix (16:12 - identify all movies that a user has rated. Then select a movie to predict, that movie's weighted avg. will be 1.00. Then identify all movies similar to that selected movie by using Pearson correlation then compute similarities between rows. When you find the similar weighted averages, to 1.00, use the equation, @ 19:13, to predict the rating for that original specified movie.
Item similarity is more meaningful because items generally belong to a small set. Users tend to have varied taste.
Key Points
Centered Cosine Similarity/Pearson Correlation (for user-user collaborative filtering): 7:18 - 11:13 Rating Predictions: 11:14 - 13:56 Item-Item Collaborative Filtering: 13:57 - 19:14
Citation
Repo link (optional)