Where do we see recommendation systems
Personalization is changing our experience of the world
The film is recommended
recommendation
Music to recommend
A friend recommended
Drug-target interaction
3. Recommended classification model
3.1 The simplest method – popularity
3.2 Classification model of Solution 1
What’s the probability that I’m going to buy this thing
Limitations of classification methods
4 Collaborative Filtering
Solution 2: Collaborative filtering
Co-occurrence matrix
Use the co-occurrence matrix to make recommendations
5. The influence of popular goods
The co-occurrence matrix must be normalized
Normalized co-occurrence matrix
similarity
- limitations
A weighted average of goods purchased
- limited
7 matrix completion problem
Solution 3: Discover hidden structures by matrix factorization
The movie is recommended
Matrix completion problem
8. Make recommendations based on the characteristics of users and items
Suppose there are d themes for each user and movie
9. Use matrix form to predict
Discover hidden structures by matrix decomposition
Matrix decomposition models: Discover topics from data
Limitations of matrix factorization
11 Put it all together: feature + matrix decomposition
Synthesize features to discover topics
Hybrid model
12 Recommended system performance measures
The world of baby products
A subset of items that the user likes
Why not use categorical accuracy
How many favorite items are recommended
Which of the recommended items do users like
13 Optimal recommendation
Maximize recall: Recommended for all items
How accurate are the results?
The optimal recommended
14. Accurate-recall curve
Which algorithm is the best
15 Summary of recommendation system
- Some features, such as product/user ID feature pairs, are selected from the table of customer product rating in the training set. The goal is to predict that these users will give corresponding ratings to these products. Therefore, the rating of user ID on product ID is our target Y hat
- Machine model matrix decomposition model,w hat is the prediction parameter
learned
