Recommendation System
Recommendation system is a system to recommend some content based on user preference.
Recently, I have some interest on a such system, since I remember some experience on having hard time finding similar anime to the one I just finished. While the recommendations on the Internet are not reliable and video website can only recommend those anime they have bought, if there is a specific anime recommendatioin system, how good it would be!
To start this project there is a lot to learn since I have only a little experience on machine learning and using models.
Here is the basic process for the project:
- Decide the architecture
- Collect data
- Clean the data
- Develop the model
- Implement the model
- Train the model
- Do evaluation
Need to learn some basic knowledge on machine learning.
Regression line: use to show the trend of the data
R square: error measurement
r^2 = 1 - Square Error of regression line/Square Error of mean line
I decided to gain some knowledge first through kaggle. Will come back soon!
I learned some high level ideas on recommendation system through a youtube video.
Here are some notes:
collaborative recommender: find similar user and introduce what he likes (through matrix completion)
content based recommender: find other similar contents u liked (use embeddings: leverage content/meat-data to embed into a low-dimensional space)
Hybrid model: two tower architecture
Architecture:
I decided to use the two tower architecture combining collaborative recommender and content based recommender.
Data:
I will use a myanimelist data set from Kaggle.
UPDATES:
In the end, I chose a different recent data set for the model.
I separated the TV anime and Movie anime, and build two model using pytorch.
Now I am working on intergrate the model into a web application!