Scenario gaming #1637
Scenario gaming #1637
Conversation
miguelgfierro
requested review from
gramhagen,
anargyri,
loomlike and
wutaomsft
as code owners
|
|
||
| ### Next best action prediction | ||
|
|
||
| An interesting scenario is next best action prediction. In this scenario, what is recommended is the most beneficial next action for the player. From the technical point of view, this can be implemented using collaborative filtering algorithms, such as [SAR](../../examples/00_quick_start/sar_movielens.ipynb), [BPR](../../examples/02_model_collaborative_filtering/cornac_bpr_deep_dive.ipynb), and [NCF](../../examples/00_quick_start/ncf_movielens.ipynb). |
There was a problem hiding this comment.
Isn't this task closer to reinforcement learning rather than CF?
There was a problem hiding this comment.
Good point, there is a way to do next best action in RL, a lot of people using MCTS. This is done in AlphaGo, AlphaZero, etc. But in that case, the selection of the optimal next action is automatically taken. If we pose the problem as a reco system, where actions are treated like items in the typical CF setup, then the user can choose among a set of actions.
There was a problem hiding this comment.
I meant something about the state i.e. in CF there is no state. But here is the current state of the game relevant for which next actions to recommend? "Next" implies that you have at least some info about what has happened before. With CF you would always recommend the same action, wouldn't you?
There was a problem hiding this comment.
The way I see it is like a session based recommender, but with actions instead of items. The info that has happened before are the previous actions, the next recommended action is an option that the user chooses among a list of actions available.
|
I'll merge this now, we can iterate over the content if needed |
Description
Add scenario for gaming industry
Related Issues
Checklist:
staging branchand not tomain branch.