Show HN: Collaborative RSS Reader

Brilliant idea! this is the web3.0 which should be focused on.

The popularity based recommendations are only shown to brand new users - so they have some content to rate.

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.

How did you come up with this mechanism? It's very clever, simple yet powerful. Once I saw I could associate upvotes with channels, I realized how useful this could be! It's like making your own subreddits that fill up with content automatically. I think after you reach a critical mass of active users, this could really take off.

From the FAQs on https://linklonk.com/about: "No, at the moment I don't have plans to open-source it."

When I find an article that is particularly insightful or informative I often wonder who are the other readers of this article, how did find it and what else did they find interesting? LinkLonk is a way to connect me to the other readers.

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.