Hey everyone! I'm working on an app that hosts a ton of videos across various categories, and I'm concerned that users might get overwhelmed by the content. I'm looking for advice on the best way to implement a recommendation feature that can grow with us, especially since this is my first time setting up something like this in a production setting.
I don't have any experience with machine learning, but I was considering creating a chatbot that users could interact with to get recommendations based on our content library.
I have a couple of questions: 1) Am I missing out on something significant in terms of quality and cost by not developing my own ML model? Would that impact scalability? 2) If I decide to switch from the chatbot model to a strategy where recommendations are generated automatically, like how YouTube does it on the homepage, would that dramatically increase costs? Or is that approach still more economical than maintaining a trained AI model?
We're currently managing around 1,000 videos and have over 5,000 active users, and those numbers are increasing fast!
4 Answers
A solid starting point would be to analyze what videos each user has already watched and categorize their viewing habits. Keep track of time spent on different videos and the order they watch them in. You could then recommend the top three videos on their homepage based on that data. This setup would be lightweight and effective.
Honestly, using a machine learning model might be overdoing it for your needs. If you implement decent tagging and track a "hotness" score based on user interactions, that could be sufficient for your user base. If you decide to use ML, maybe consider a simple classifier to figure out relationships between videos instead of building a complex model from scratch.
Thanks for your insight!
You might want to look into graph databases for tracking relationships between users and video tags. As you accumulate more views on specific tags, you can boost their overall score. This method can help you create a more tailored recommendation engine over time without relying on a complex AI solution. I'm not really a fan of the one-size-fits-all approach, as those can be tough to maintain.
Great point, I'll explore graph databases!
If you want robust search capabilities without diving deep into AI, definitely check out ElasticSearch. I’ve used it and while there’s a bit of a learning curve, it really allows you to fine-tune search results and relevance. Plus, it scales well, so handling 1,000 videos is nothing for it—you could easily expand to 10,000 or even 100,000.
Thanks for the recommendation!
Awesome, I like that idea!