On Engagement Metrics

At present, our discussion page lacks any statistical data about the thread, a feature that is pretty common in most of social networks or forums. These platforms typically display a total count of replies, boosts, likes, and participants.

Take Discourse as an example, a platform with a great UX for productive conversations. Discourse displays the list of links shared in a post and provides a statistical box that details thread engagement. This includes also an approximation of the time required to read through the entire thread, if significant.

What are some meaningful statistics we must/should incorporate for the Bonfire 1.0 release? We are seeking to find statistics that can effectively foster engagement in discussions without solely promoting addictive behaviors. Any thoughts or suggestions ? #bonifre_feedback

Sorry for bringing this thread back from the grave 3 months later, I just found it again from an unchecked notification from months ago 馃槄

I just wanted to say that I really like the current implementation as of this version (0.9.6) with showing the number of replies, and also the number of participants. The "Last Reply" is also something that I would never have thought of myself but I think is incredibly useful.

I do not have anything substantial to add, other than just saying I appreciate your approach to community suggestions and I think you did a great job with the UI/UX on this feature.

+ -


With this entry, <1 day of engagement here. Enticed by some posts by https://xn--ocane-csa.fr/ on #Mastodon and their website. I've only just started to read https://bonfirenetworks.org/docs/ and https://www.w3.org/TR/social-web-protocols/

See document: https://www.platformaccountability.com/proposal, then scroll to the bottom. It shows some suggested measurements and some objectives for those measurements. Some for within a post/thread and some situated elsewhere TBD. The white paper contains more detail.

So sorry to be commenting without completely reading the specs first. Understood this particular comment period was for 2 weeks, and unsure I could finish and grok the specs in time to have a worthy basis for comment. I couldn't find a reference to the platform accountability document anywhere in the #Fediverse yet (but that be my weakness in search skills here). Wanted to be sure this work was known. Seemed like there were at least overlapping goals in the bonfire measurement goals noted here and the platform accountability goals.

+ -

@ivan Maybe show the number of flagged/reported posts?

Usually when I see many posts being flagged it's a sign of the thread going wildly off-topic or even attracting bad actors. I then stay away from that discussion. However, e.g. in Discourse I won't know until I have read through the posts. Having that metric on the top post would help me spend my time and engagement elsewhere.

Thinking of risks, maybe showing flag stats would attract more of the same? I would hope not in an otherwise functional community, but some people seem to gather around car crashes or houses on fire.

+ -

@lne @ivan This does make me think though that maybe showing some metrics (like boosts and likes) as an aggregate for an entire thread (instead of for an individual post or comment) may be interesting indicator while being less connected to ego and addiction?

@mayel @ivan Sounds good, I think that would help me decide on which threads to read through.

I assume there still has to be some kind of clickable "You liked this post" indicator on posts to allow people to "undo" a like.

+ -

@lne super interesting, wondering if it is a valid metric though and in which ways it may have privacy implication ? A thread may have a lot of flagged activities because it talks about a complex topic (covid, politics, religion, gender ...) and users want to properly moderate it, on the contrary a thread without flagged activities may be just not moderated at all and being full of trolls ?

Maybe we can show a label only when the amount of flags exceed a certain amount (eg. after 30 flags on the whole thread) - as an indication 馃

> which ways it may have privacy implication ?

My intuition is that the number of flags in a thread would have the same privacy implications as any other aggregated metric (#posts, #likes, #people,...). Discourse folds flagged posts, which should have about the same privacy implications as showing the number of likes.

Or am I missing something?

> wondering if it is a valid metric though ... A thread
> may have a lot of flagged activities because it talks
> about a complex topic (covid, politics, religion, gender ...)

I agree, many flags could mean different things depending on the topic and the community.

On the other hand, couldn't that also be said about other metrics to some extent? I mean in the sense that before jumping in, I need to use my judgement based on general context (e.g knowledge of the community) and topic title, avatar list, available metrics, etc.

Interesting things can be learnt from threads on hot or complex topics, or where a different set of people engage, and I don't mind being part of community moderation. It's just that being the umpteenth person attempting to get a derailed thread back on track feels like I'm wasting my energy.

In a functional community this is just the odd thread, though, so probably less of a concern for the kind of thing Bonfire aims for?

+ -

@mayel @ivan
Quantity of comments is good and necessary and has already been discussed.

WDYT about including the "position" in the discussion?
It helps to understand the present comment/post in context.
This indicator has a simple structure: P/T where:

  • P is position and
  • T is total number of comments.

I know it's not that easy to count (do we count bifurcations and/or other open lines of discussion?) buy it gives some idea.

It is not the same to see "23 comments" that "comment 21 of 23" or "comment 4 of 23".

Ideas, critiques, hurrahs, anguished screams or other reactions to this? 馃槈

+ -
+ -

@ivan @mayel @edumerco


Perhaps there is an idea for an algorithm in this report.

Here's another example of quantifying elements of a thread with real world example: https://sites.dartmouth.edu/learninganalytics/2016/01/26/the-application-of-social-network-analysis-in-canvas-discussion/

From a data portability perspective, when an actor moves themselves (and their past posts) to a new instance, each post in a thread can contain a vector of the original posting "punch" based on metrics. When the actor's past posts are discovered by the actors of the new instance it may spawn a new thread graph. This implies these "punch" metrics are supported in the new instance. Maybe as a microsyntax?

@mayel @ivan @edumerco

Another thing about the metric... trust and robustness to editing and deleting. If the post is moved by the actor(owner), trust in the "punch" allocated on the initial thread and robustness to subsequent editing (before or after move) would need to be incorporated in the "punch" metric.

Just arbitrary, calling it "punch". There is probably an official name in some standard (that I haven't found). I think it is equivalent to the flame icon mentioned elsewhere. To first order, ActivityPub (AP) spec seems to put it in the "microsyntax" vocabulary with "may" connection with other AP implementations (?languages?... Mastodon, et al.). It doesn't seem to be an ActivityType such as Like (in the AP spec). I looked in the Bonfire documentation, but I don't know how to find how the "flame" maps to AP.

@ivan @mayel
A few possibilities (starting from my total ignorance of the topic) to ignite the discussion:

  • take the longest linear thread that the post is in and count it's position there and that whole thread number as length. Easy to do, not very close to reality, but a simple possible way to start.
  • take the total number of replies that open from the original and use the timestamp of the current comment as the position. This ignores the bifurcations and uses time as linear count, it is more representative of the whole volume of the discussion and not really a position, but a reasonable approximation to the order of the posts.

What other algos can we think?

+ -

for sure what helps me to engage is the comments number. that works for me more than any other stat, cause i see there is a discussion i can learn from and eventually engage in.

anyway maybe you could put a little counter on or under every of the buttons that are already there?


  • under or on the right of the reply button, the number of the replies already there
  • under or on the right of the repost button, the number of the repost already made and you can see who did it by clicking on it
    etc etc
+ -
+ -

@zabbeer @ivan @lefterino Yeah I think showing the number of comments (both in feeds and at the top of a discussion page) would be useful. Maybe better to not show the number of boosts/likes by default though, and either make that a setting you enable or a toggle/button/menu to explicitly choose when see them.

+ -

now that i think of it, can bonfire not follow mastodon on the ''like'' = star thing?lol

everytime i click on it i have years of star= favorite bookmark taught by internet in the brain

a bonfire is a fire sooooooooooooooo, maybe a fire reaction? no likes, only hyped fires 馃榿

+ -

@ivan @lefterino okay awesome! when y鈥檃ll start adding sound effects鈥lick it has to be some noise like a short fire crackle or something. Right? Something low and not too annoying鈥hat can be turned off.

+ -

No doubt Discourse's stats are awesome!
I would say it would be nice to have those avatars at the bottom of people who have engaged in the conversation - showing first the people you follow or you mostly engage with and an option to expand to get the whole list of people who interacted with the post. Maybe -this could be controversial- an AI-derived icon next to people's avatars in the list indicating if they agree or disagree with the original post.

wdym with agree or disagree? i mean, the ''ai'' would first check the post and then try to understand if my comment agrees and if it does the icon is different?

@jl I also agree that participant avatars are a relevant info to show 馃憤锔 .
Re. AI: we actually did some very first experiments in the field of sentiment analysis , but to properly integrate AI in bonfire we need to do some work on the infrastructure side, to avoid putting any strain on the (typically small VPS) instance鈥檚 server