Show HN: Collaborative RSS Reader

At this scale the project doesn't cost much to run. The best motivator for me to develop it further would be your feedback and any help spreading the word to bring in more users. After all, in order for the "collaborative" effects to kick in we need more users.

A new user starts off weakly connected to all other users and to feeds the existing users are connected to.

But as you submit or upvote links to content you liked - your recommendations will be based on your explicit upvotes, and will be less popularity based.

> aren't you recreating some of the same issues with social media?

1. The algorithm.

Social media algorithms are black box ML models that optimize for time-spent on the platform (ie, ads watched). LinkLonk also uses an algorithm, but there is 0% of machine learning. All it does is keeps track of how useful the past recommendations of every RSS feed and every user have been to you. You define what is useful to you based on what you upvote. This makes it so that each user can make decisions of what they want to optimize for. Social media algorithms make these decisions for you.

2. The types of actions

In social media we have two types of actions:

- Inbound: follow/subscribe - they affect what I see as a user. The target of these actions is myself in the future.

- Outbound: comments/likes/dislikes/retweets/etc - they affect what other people see. The target - are other people.

The problem is largely with the outbound actions - social media give us tools to influence other people and we use them for that. In other words, they give us megaphones and we each crank up the volume of our own megaphone to the max. There is little incentive to not set the volume any lower than the max.

I understand your perspective - traditional RSS readers don't have any outbound actions. You only decide for yourself who you trust to get information from.

And your concern is that LinkLonk provides tools to influence other people and that these tools will be inevitably abused.

What makes me hopeful about LinkLonk is that it does not have separate outbound and inbound actions. There is only an upvote which is both inbound (determines what your future self sees) and outbound (it affects what other people see). So you may want to be more careful about what you upvote - because you don't want your future self to receive useless stuff. For example, maybe you would be less likely to upvote an article without reading it purely based on the title that says "Look how terrible the other side is".

3. Proof of work

The amount of attention someone gets from you is directly proportional to how useful their past recommendations have been to you. In other words, to get your attention they need to prove to be good curators of content for you. They cannot fake it by upvoting the most popular content. That’s because you connect only to those users who have upvoted the content you like before you did it - ie, before the content became popular.

By the way, you can group your upvoted content into channels. For example, if you are interested in tech news and food, you can create a channel for each interest. And then people who like you "food" upvotes will see your other "food" upvotes but not your "tech news" upvotes and vice versa. This would create more coherent groupings.

More on channels in this FAQ post: https://linklonk.com/item/2369676162410512384

If such RSS feed includes content from all sources that you are connected to - then your RSS feed would have too much noise. The only way I can think of converting it to an RSS feed is if you choose how many top recommendations you would like to see in that feed per day.

This is how the email newsletter works today (you can subscribe here: https://linklonk.com/profile) - you choose when (daily/weekly and at what time) and how many top recommendation you want to delivered to your email inbox. And you can rate items from the email.

> Can it be seeded with known liked feeds or content?

Yes, you can seed your recommendations - simply submit (https://linklonk.com/submit) links to content that those feeds posted or directly submit links to those feeds. If you use RSS readers then you can the list of feeds you are subscribed to as an OPML file. I haven't built an automatic way to import OPML files, but if you send your OPML file to me by email, I would be happy to import it for you.

> And can you 'pin' liked feeds in some way, to say 'I always want to see content from here'?

To soft-pin a feed that you liked you need to upvote content from that feed. That will cause your connection to the feed to be stronger than to others and items from that feed will show up at the top for you.

There is no explicit pinning functionality - something that would effectively set the connection weight to infinity. The value proposition of LinkLonk is that it keeps track of how useful the content from each feed and user has been for you and prioritizes new content accordingly. I know this is very different from a traditional way of subscribing to RSS feeds, where you are either subscribed or not (boolean). On LinkLonk your subscription has a weight (float). In the days of Google Reader, there was something similar I think if you sorted your content "by magic". Only on LinkLonk, there is no magic - the mechanism that connects you to feeds and users is transparent and controlled by your upvotes.

> Is this a reader, or another source? It is a different kind of reader - one that automatically manages your subscriptions.

I guess it could be trivialized to the good old "Other people who bought this also bought that" but for articles. But that was not the source of inspiration.

The algorithm itself is inspired by Personalized PageRank (which Pinterest used to use 3 years ago: https://blog.acolyer.org/2018/05/23/pixie-a-system-for-recom...). Essentially, you do a walk on the graph that has sources of information (RSS feeds, user channels) as nodes and upvotes as directed edges. The walk starts at you => into one of the items you upvoted => into one of the sources that upvoted it before you => into one of the other items that source upvoted. You sum up all the probabilities to end up in each item after these 3 steps - and that gives you the ranking score of the item.