Recommendation System

[ahle]

Traditional recommendation system based on the user rating on the similar items.

So, we can predict the user rating on an unrated item based on the his rates with the similar items.

source : http://fr.slideshare.net/VikrantArya/recommendation-system-33379953

In the trace context, the item here can be an episode, an obsel, or something. But how we know the user prefer (like) this episode or this obsel ?

Some works need to be done before implementing an recommendation system:

• similarity
• rating system
• similar user vs all user vs target user
• prediction
• predicted objects
• data vs modeled trace

[amille]

Il faut sans doute que tu fasses du ménage dans tes "problèmes à résoudre" (issues). Aucun n'a été résolu depuis le début ?

[ahle]

The recent learning algorithms:

• Reinforcement Learning
• Deep learning
• Q-learning

Todo:

• Apply the learning algorithms with small data
• Find the appropriate learning algorithms with bigger data

-> build dataset from trace