I was watching the video yesterday and thought of this idea.
Lets compare:
Youtube know users. It will recommend videos for users.
What Devto give me now? It always give me new posts.
Now recommendation is everywhere. Search age is passing. Now is recommendation time.
In China there is a site toutiao.com(Today news you care using AI based recommendation) is becoming popular.
Any idea?
I agree with Max, while also thinking we can get a lot better at recommendations based on context and basic data we share (what tags you follow). Our current system is fairly naive. We don't want to start gathering personal data (which I don't think we need to provide the right experience), but there is definitely room for improvement.
So...
Thanks Ben, just a gentle idea from the experience from using Youtube.
I know that is not easy to do. But it's really worth to care.
Based on my limited knowledge of the topic, I’d say that the most important requirement for proper recommendations is not personal but behqvioural data: who reads what, who reacts to what etc.
A dataset relating text to chance of enjoyment already exists in the form of likes. No apparent need to gather any extra data.
It definitely does not need a recommendation system like Youtube, I don't want to be suggested the same five articles as everyone else that I have seen multiple times already 🙃
I think the current suggestion system is good enough. Making a suggestion algorithm that works well for everyone is a task that multiple organizations are collectively firing billions of dollars at. I don't expect dev.to to be able to nail it.
Agree. and the one thing I hate is the click-bait title that I've already read it in the past.
I like the current timeline today. and finding another click-bait popular posts eases me to read more in this community.
Please NO. Recommendations make you live inside a bubble of your own already known interests, and I'm using dev.to for the very opposite reason: I follow the global RSS feed an I pass through EVERY post EVERY day; if the argument appeals me I read, otherwise I swipe in a few seconds.
I don't want something that feeds me with what it thinks it's ok for me. I'm not an infant and I don't need breastfeeding.
Space on screen is finite. The more related content you show, the less proper content you have.
I think the finite space argument could actually go in favor of having a recommendation system in the end. Imagine that in 5 years dev.to is so big that it's impossible for anyone to go through "EVERY post EVERY day", unless they dedicate the entire day to it. In such a case, finding the posts that you actually care about (or simply would like to read) would be really difficult and people could start turning away from the platform. Unless it found a way to show reasonable amounts of interesting content...
Still, there's a difference between the user choosing which content they want to be fed (in terms of following tags and defining weights for them, currently) and the system choosing which content to feed to the user (like extrapolating from previous reads and other users with similar behaviours).
I'm fine with things like a list of similar/recommended articles at the bottom of an article, but I wouldn't want the content in my front page to be even more selected by bots.
This is a trivial example of the well-known problem of balancing exploration with exploitation, and has an equally trivial solution: just add some random stuff to the recommendations.
I would also recommend against this, partly because I don't think Dev.to is at a size yet where more personalized recommendations would make much of a difference. This community is still small enough that I will frequently visit all latest articles and not feel overwhelmed, and from there get to know other community members I want to engage with.
In addition, @ben and his team do a great job of publicly praising great contributions and new users, in addition to highlighting popular tags and new discussions in the feed sidebar, which really helps ensure no one is lost in the crowd and helps me follow new, great members.
I think a YouTube-style recommendation system might be effective several years down the road as Dev.to grows in size, but I just don't think it's needed yet.
If there is a recommendation system similar to this based on our tastes, we could pair it with a "tastebreaker" recommendation system. This would basically look for posts you wouldn't ordinarily read due to your tastes as a way to expose you to new writing and ideas.
Implementing a recommendation system like YouTube on Dev.to would require advanced software development techniques, including machine learning, user behavior analysis, and scalable backend architecture.
Nope I think the current recommendation is fine. YouTube are able to provide recommendations is because they gather personal data. I do not want to look at unrelated articles besides the topics I subscribe.