TrustGraph integration

[mattyg]

Originally by @mattyg , Collab edited by @harlantwood, @dauphin3, @jost-s --

@harlantwood & @dauphin3 are working on integrating TrustGraph 🎉 in #104

Collecting thoughts -- about the initial integration and beyond -- in one place.

Overview

Following

• Instead of "following" an agent being a binary action (either followed or not followed), with trust graphs we can add granularity and context to a follow.
  • Each follow will include: a topic (the type of content that you trust them on) and a trust rating (decimal between between -1.0 and 1.0 of trustworthiness).
• when to exclude mews -- UI should be very clear -- eg a slider like: [Hard Pass]---[Meh]---[Awesome] which mirrors the underlying scale: -1.0 to 0 to 1.0.

Feed

• Split feed UI into 2 tabs -- Chronological (the current feed) and Recommended (TrustGraph powered feed)
• in recommended tab -- needs to be weight and recency combined -- don't want to see very old highly weighted at top
  • could be a simple linear function - eg 2 week window - or window varying per user based on usage patterns
• could have visual tag showing eg ( holochain, 98% ) next to mew to show why I'm seeing it -- ie the properties with which I tagged its author when I followed them
• ideally if I've scrolled one tab and go to the other I don't see mews repeated
• Variable depth of network
  • Display only agents within a trust graph depth of 5 of my trust network (for discovery)
  • Display only agents within a trust graph depth of 3, on the topic of "cooking" (for finding highly trusted chefs and recipes)
  • Display only agents with a trust graph depth > 5 and < 20 (for discovery of agents far outside my trust graph perspective)

Topics

• since we need a concept of "topic" that we trust an agent about, we should incorporate that into the UX. Perhaps by tagging mews with a variety of topics, leaning on #hashtags as topics, or using machine learning to identify commonly used words & phrases
• Display "word could" of topics on a user's profile page. Can we come up with a novel modern UX to bring back word clouds?

Implementation

• In addition to our current "follows", we optionally add TrustAtoms which can be thought of as "nuanced follow information" -- eg ( "cuisine", 99% ).
• On "unfollow", we will delete the relevant TrustAtoms (depends on Allow deleting TrustAtoms trustgraph/trustgraph-holochain#67)
• In "Feed - Recommended tab"
  • We query all TrustAtoms created by the Agent
  • We filter for those TAs above a weight threshold (nominally >= 0)
  • We follow those links to get their targets (followed agents)
  • We fetch posts from those agents which mention the relevant hashtags (do we store links for hashtags in DHT currently?)
  • We order these posts according to some combination of: weight of tags and recency
• Other ways we may want to retrieve TAs (TODO add UX use cases)
  • for all of: follow / following / my_followers / my_following zome calls
    • we could add a boolean include_trust_atoms argument
    • we could allow filtering by topic

[dauphin3]

doesn't it only go to 7 degrees of separation haha

love the idea of word clouds.. I think its good to create a matrix of tags for each "topic" instead of trying to find an umbrella term, so that topics dont converge on a single idea but allow for multiple meanings linked to components of concepts/categories

[harlantwood]

Thanks @mattyg for writing this up! Great summary.

[harlantwood]

@mattyg FYI we're treating the “description” field as a bit of a wiki to ideate on the features and implementation details.

[harlantwood]

@jost-s and I did a pass on design and implementation of relevant features -- notes integrated into Description field above.

FYI @mattyg @artbrock @bierlingm @dauphin3

[bierlingm]

I like it. Anything helpful we can contribute, or just an FYI?

[harlantwood]

I like it. Anything helpful we can contribute, or just an FYI?

Thanks for asking @bierlingm -- feedback and ideas or concerns welcome at this stage.

I also made a more tightly scoped description of the initial PR here: #104

[mattyg]

We decided to remove the native "follow" feature entirely and instead just use trust graph for follows.

• If you're not following somebody, the UX in their Agent Profile Popup will be the same as before: just a single follow button.
• If you are following them, the UX in their Agent Profile Popup will show the list of topics and trust weights (if specified).
• By default, a "follow" creates a trustatom for the "catchall other" topic at 50% trust

[mattyg]

Some more thoughts from our meeting today about trust graph UX:

• We shouldn't have to type in the tag we wish to follow someone on manually. Even with typeahead search this is an odd clunky interface.
• Display a tag cloud of hashtags / cashtags used in agent's mews
• User can adjust the trust weight slider for tags within the tag cloud UX

I definitely understand if this doesn't happen in the first pass of trustgraph integration, but if so let's make another ticket for it.

[mattyg]

Another idea from our meeting today:

• ability to adjust your trust weight for an agent directly from a Mew they published -- perhaps in the agent profile popup

[mattyg]

Reflection from meeting today:

• Trust weight should be very coarse -- i.e. "none" (0%), "partial" (50%), "full" (100%) -- making it too granular (i.e. 0-100% slider) feels unnatural for how we think