Nostr Web Client

We need a way to do algorithms which learn from user interactions. Ideally an open system, that users can choose where to give their interaction data and what their algorithms are optimized for. But we can't just put our heads in the sand and ignore their existence. All of the A/B studies where users were given chronological timelines show they'll switch their time to other apps / services that do provide an algorithmic timeline.

The question is how do we do it in an open, nostrway...

If we don't then we get a social network owned and run by Sam Altman... which honestly is terrifying. He'd make Elon and Mark seem like balanced stewards in comparison.

https://www.theverge.com/openai/648130/openai-social-network-x-competitor

ManiMe 6mo ago

nostr:npub1wmr34t36fy03m8hvgl96zl3znndyzyaqhwmwdtshwmtkg03fetaqhjg240, we already have an extensible recommendation engine ready to be built upon in exactly this manner. It’s called GrapeRank nostr:npub1u5njm6g5h5cpw4wy8xugu62e5s7f6fnysv0sj0z3a8rengt2zqhsxrldq3 designed the algo and I developed the library.

At its core, GrapeRank is a WoT engine. It generates a weighted list of “influential” users from the perspective of a single “observer”, by ingesting and interpreting any kind of content (follows, mutes, reports, ect..) from across the network.

These multiple and customizable “recommended user lists” can then be used to build content feeds … by identifying the authoritative or desirable users on a given topic.

The point I’m trying to make is that the open and extensible recommendation engine we’ve already built will bring Nostr closer to having customizable feed algos… and that these feeds must be built on a user customizable WoT engine. We have this.

I am actively looking for funding to develop GrapeRank further.

https://github.com/Pretty-Good-Freedom-Tech/graperank-nodejs

Reply to this note

Please Login to reply.

Discussion

rabble 6mo ago

I think we need to base it off of user interaction and activity not follow graphs. But you can build it with the follow graph and lists and see how it works. It just feels like the way things were done a decade ago. Which will work for somethings fine.

ManiMe 6mo ago

nostr:nprofile1qqs8d3c64cayj8canmky0jap0c3fekjpzwsthdhx4cthd4my8c5u47sprpmhxue69uhkwun0w4c8xtnxd9shg6npvchxxmmdqyf8wumn8ghj7ur4wfcxcetsv9njuetnw6p4df thanks for taking a look. Did you notice that GrapeRank has NOTHING explicitly to do with follows?

Use GrapeRank to make a list of users matching ANY interaction criteria. It’s a library waiting to be extended. Heres how easy it is … let me help you extend it further :

https://github.com/Pretty-Good-Freedom-Tech/graperank-nodejs/blob/main/packages/ptotocols-nostr/index.ts

rabble 6mo ago

I’ll take a deeper look

ManiMe 6mo ago

I’m here to help.

ManiMe 6mo ago

I’d be happy to run a demo presentation for you or your team.

david 6mo ago

We don’t have to choose between follows vs user interactions vs something else. Our trust graphs are ultimately going to incorporate all sources of data available to us: follows, mutes, reports, reactions, replies, zaps, etc.

The key to making this work is a step that I like to call interpretation: find some raw data sources (like replies and other user interactions), execute a script that translates the raw data into a standardized format that is ready to be digested by your trust score calculation engine. This is how GrapeRank works.

My goal is to create personalized WoT relays that use GrapeRank to calculate contextual trust scores, with interpretations tailored to fit your personal preferences and beliefs.

david 6mo ago

Just to clarify: “standardized format” is internal to your WoT relay, not necessarily standardized across relays. Meaning: All interpreted data gets transformed into the same format inside the same relay. But the internal standard that your WoT relay uses does not necessarily have to be the exact same internal standard that my WoT relay uses.