Topic Summaries: when the conversation is over


(Chris Hanel) #1

Since Discourse first appeared online, my interest has primarily revolved around the concept that a forum, by definition, exists to facilitate a discussion that, in the end, will generate valuable content for community members and visitors alike. The natural trade-off is that the format of the content forever remains in a state of conversation: A collection of posts and replies with a horrible signal-to-noise ratio, the noise providing nothing but some context to the info that inevitably does matter when someone digs it up looking for the answers they seek.

Discourse has already made steps to take on this problem from one of two different angles: Either by attempting to prevent an increase in noise (trust level moderation, similar topic feedback), or by helping a user navigate through it (improved search, ‘best of’ topic sorting).

However, I’m curious if there isn’t an opportunity to actually tackle the problem by actually reducing the noise in a non-destructive manner. We’ve talked in the past about allowing a post to be flagged as ‘best answer’, with some good back and forth as to the pros and cons, but I realize that there are frequently times where such a feature is insufficient to provide the needed info without further searching through replies.

So, a series of thoughts:

  • What if Discourse offered features similar to, say, Storify, and gave trusted users the ability to quickly create a summary of a topic (or multiple topics), adding additional context where necessary and automatically embedding/citing/linking excerpts?
  • What if such a summary could earn sufficient approval, and a visitor reading a topic would see the summary first, and could still scroll below it to read the full topic for themselves, should they wish?
  • What if such a summary could be pushed into a blog feed, motivating site owners/admins to summarize topics by rewarding them with easily generated (and useful) site content that attributes contributors without any additional effort?
  • What if users who are cited were notified, and allowed to weigh in as an ‘aside’ to accompany the summary?

This has similarities to the past proposal of offering a “wiki topic” with a space at the top for open editing, though I think I prefer this alternative to the open-endedness of free-form content creation.

Lots more to say on this topic, but interested to hear initial reactions.


When does Summarize kick in, and what can I do about it as the user who gave the definitive answer?
(Sam Saffron) #2

I like this idea and have come across it before in a few spots, but I think it is a tad premature. First we need the ability to wiki topics.

For example, the howto category could be a “wiki category” meaning that any trust level 1 users (configurable) can edit the OP and improve it.

Just building that feature alone is a big can-o-worms:

  1. You need a more decent diff function than what we have
  2. May need better notifications around this
  3. Need the ability to revert changes in a simpler way
  4. We need to amend the bumping logic to some degree.

I think getting wiki going is the best first step here, its a simple concept to explain and is much less work than getting into the summary game.


(Erlend Sogge Heggen) #3

This is a prime example of “easier said than done”, but I’m rooting for it none the less. I’ve wanted this for my own developer-centric forum for a very long time, since more often than not the answer is in a string of back-and-forth between 2-3 developers rather than a single post by one author.

The “best of” feature does a pretty good job of alleviating this problem already though, and definitely merits further exploration of it capabilities.


(Kim) #4

That could be very interesting for my community. We use our discussion forums to collaborative write novels and novelettes, the exporting process back to MS Word once the writing is done is a nightmare. :smile:


(Jeff Atwood) #5

Why not just automatically force topics into “best of” mode (or an enhancement thereof) once they get to a certain size?

I like this idea but it is also 99% what “best of” is trying to do – make the topic summarize itself!

Because forcing someone to write a high level summary of the topic is work that just won’t reliably get done. Make computers do it.


(Chris Hanel) #6

No, agreed, “Best of” has a lot of overlap here. In fact, I will go on the record here and say that my past skepticism about the feature has largely been proven wrong, and that my doubt was fueled by being averse to systems susceptible to circlejerk/hivemind type scenarios. I still believe that there’s an element of it present and always will be in a system that is informed through liking/upvoting, but not nearly as much as I feared.

I look at a feature like Thread Summaries as offering two positives that “Best of” cannot reliably deliver on, both of which are important to communities where the admins and more trusted members are seen as educators and the community exists for members to learn from each other.

The first is offering a method of curation that’s more elegant by providing an opportunity to truly ‘wrap a bow’ around the thread and make its digestion less… cinema verite?* There are many situations where a thread evolves to the point where the OP doesn’t provide a natural starting point for reading the thread after the fact, the content has a flavor of ‘you had to be there’ where participants in the conversation assumed another piece of information as common knowledge, thus making the thread cryptic to outsiders, or a seemingly innocuous post that wouldn’t be displayed in a “Best of” thread provides a vital data point that other posts pivot on. These are the kind of situations where I would find it incredibly useful to be able to summarize at a human level.

The second benefit, I believe, is related to this point:

Thread summaries give the ability to site admins to easily provide meaningful content to non-forum members. I’m a member of multiple communities where a forum exists alongside a blog on an educational site. The admins provide tutelage, answer questions, and write blog posts and tutorials, but frequently the perceived ‘secondary’ benefit of the community (self-educating through collaboration) becomes too relied upon to provide value when the educators themselves don’t have the time to create original content. In this situation, thread summaries can provide a huge incentive to all parties: Educators can simultaneously showcase the value of the forum and have an easy method of creating meaningful content for the website, users are motivated to contribute more and earn some time in the spotlight through attribution, and of course the obvious benefit of having a summary visible in the original forum thread when someone searches for it there.

I understand that not every community has this motivation or this inherent need filled, so maybe something along these lines is better for discussion as an extension rather than for core, but I do believe that something like this, if done right, would be a big step forward for making a forum more useful over time… and, of course, this means that the corrolary is likely true as well, and that a half-baked implementation could easily destroy everything. :smiley:

*We beat up people for making analogies at this level of pretentiousness, right? Forget I said anything.


(Jeff Atwood) #7

Couldn’t the mod edit the first post, putting a horizontal rule at the bottom, and add the topic summary right there at the bottom of the first post? Not perfect implementation wise but very much the same intended effect right now today. And all revisions are tracked by content and author…

I guess I am saying what you describe is possible already in practical terms.


(Jeff Atwood) #8

Looks like @mcwumbly was already building some manual topic summaries lately.

Hard problem.


(Dave McClure) #9

It’s not easy doing it manually either, by the way. :grimacing:


(Jeff Atwood) #10

You could compare your manually compiled summary post with what “Summarize This Topic” generates, provided the topic has > 50 replies…


When does Summarize kick in, and what can I do about it as the user who gave the definitive answer?
(Dave McClure) #11

Well, I used the ‘summarize this topic’ feature to help guide me, so I imagine there’s a pretty good correlation there :wink:


(Jeff Atwood) #12

I think the spirit of the request is on target, but the implementation is fraught with difficulty. As another person brought up in another topic, Quora also has this “summarize” feature but it is rarely used since the vote sorting (Quora is Q&A) pushes the most highly voted answers to the top. That’s not in any way a summary, but it seems to be good enough, often enough, that nobody bothers summarizing – it’s a lot more work than just voting on stuff.


(Jeff Atwood) #13

(Jeff Atwood) #14

(Jeff Atwood) #15

Reopening in case @anarcat has something to add?


(Anarcat) #16

Thanks!

I guess this topic is an excellent example of the problem that we’re trying to solve here and in a topic I opened as a dupe of this (but seems to be deleted now). Basically, there’s a fairly long conversation that happened here over the course of a year. To get a good grasp of it, there’s about 1500 words to read, which would take an average read almost 10 minutes to go through. The topic summary feature doesn’t kick in because there’s not enough juice (I guess? or likes?) so it’s unclear to someone just sifting through what actually happened here.

Having a way to either tagging a specific comment as a summary or just adding an extra “summary” field in the top post would help in better documenting discussions that happened in the past, but more importantly, actual decisions that were taken. Summarizing a conversation is also a great way to defuse tensions in difficult discussions, as it allows everyone to take a step back and look at the facts of the matter again in another light.

But I think everyone agrees better summaries would be good, just that, as you said, imlpementation is hard. I must admit this is where my ignorance of the inner workings of Discourse show: I am not a mod anywhere, just a random user, so I don’t know what’s involved in running a Discourse community. It seems to me relying on mods to write summaries for topics is asking for a lot: it takes time! Whereas anyone could (say) “ask for a post to be promoted as a summary” or some other gizmo. This could be voted up by the community or approved by the moderator or something. :wink: I am a wiki editor elsewhere, and in that culture, permissions are way more open so that kind of problem doesn’t show up, at least not in that way…

So I guess my main point was that summaries are not only useful “when the conversation is over”, but also in heated discussions, to help people figure out what the controversies are. Moderation of summaries, for me, is an implementation detail that shouldn’t keep us from considering this as an option, even if it’s just in the long term.

Thanks for the soapbox! I hope I used it well…

* anarcat steps off the soapbox :wink:


(Jeff Atwood) #17

This requires a minimum number of posts. Might be a site setting?

Problem is, a real written editorial summary is substantial work by an actual human being. Who’s gonna do that work, reliably, and forever?


(Sam Saffron) #18

Agree 100%, I do think though that in the next 2-10 years we will have very nice “giant blob of text” summarization algorithms.

I think the first way to move in that direction would be for some third party to build a plugin for it and share some examples.


(Jeff Atwood) #19

That’s summarizing one article, which is quite different than summarizing a conversation between 5-20 different people…


(Sam Saffron) #20

Agree, but effectively topics are a “story” where “sam says: X” and “bob says: Y” so similar patterns could hold. I think the quoting and replying to multiple things is the thing that would throw off current algorithms. Within 10 years there will be quite a few “magic robot sprinkles” AI out there that can summarize topics. I do not think there is anything open or closed that can solve this cleanly for Discourse today.