Traditional methods of theoretical recommendation system
Article Directory
- Traditional methods of theoretical recommendation system
- Based on the principle of content recommendation system
- Matrix decomposition of principle-based recommendation system
- Based on the principle of commodities collaborative filtering recommendation system
- Based on the principle of user collaborative filtering recommendation system
- Cold start problem
- How to evaluate the advantages and disadvantages of these algorithms as well as the performance of the recommendation system
- Hybrid algorithm
- Recommended System Assessment
- Using matrix decomposition build a movie recommendation system
Movie recommendation system works:
data is needed: a movie score sheet movie content matrix
Based on the principle of content recommendation system
Cost function
Matrix decomposition of principle-based recommendation system
Movie Ratings Table user preference matrix
for the degree of preference for each type of film has been rated by users as well as movie recommendations based on each user for each movie, each user to predict the score not seen the movie
Minimizing the cost function
matrix factorization: a matrix decomposition of movie content user preference matrix
are two methods according to the preferences, movie or similar find similar users find
Based on the principle of commodities collaborative filtering recommendation system
Similarity of goods
similar function:
predict the user to predict commodity scores
Based on the principle of user collaborative filtering recommendation system
User similarity
User rating forecast for commodities
Cold start problem
Provided that the user already exists recommend certain actions, but there is no record behavior, become cold start problem
1. User cold start: Random Recommendation
2. Product Cold start: Recommended new releases
How to evaluate the advantages and disadvantages of these algorithms as well as the performance of the recommendation system
1. Content-based:
Cons:
content analysis requires a thorough
little surprise to the user
2. collaborative filtering:
advantages:
it can be concluded product quality
You do not need to have a good understanding of professional knowledge of the commodity
Cons:
cold-start problems
Complexity will increase as users of goods and
synonym impact
will be scalping attack
Hybrid algorithm
1. Mixed Model
2 Feature Fusion
3. recommend different results in different environments
4. produce a result, a delete selection, the linear fusion
Recommended System Assessment
Under Line: Academic
RMSE:
RECALL recommendation to buy or click on the number of
probability recommendation to buy or click: Precision
Precision and recall the fusion F1: 2x (preXrecall) / (pre + recall)
is similar to a standard classification
line: Commercial
A / B testing
The recommendation system applied to different user groups
CTR
CTR degree of interest
CR
conversion experience better
ROI (return on investment - Cost) / Cost
Using matrix decomposition build a movie recommendation system
1. Data collection
Data Web site: MovieLens | GroupLens
2. Data preparation
film scoring matrix
user data on the film score
movie content matrix
movie name and movie categories of data
the user whether the film score
4. Construction of the model
code to connect: