In my last post I mentioned the possibility of learning more about the structure(s) and behaviour(s) within the dpreview forums through the process of developing a model to describe it. Well, developing models is hard-going, so I deciding to tackle it from a more interesting angle and spent some time attempting to represent forum activity graphically.
Digging a hole for one's self
How and where to start? Well, there are myriad metaphors for the concept of 'conversations', but I like to frame them as Darwinian competitions between ideas in which the 'fittest' ideas 'breed' more 'successful' responses thus carrying on the line (of conversation ;). Hence, my first attempt to create a graphical representation of forum activity involves representing a single conversation in a quasi-organic sense, to see whether it is reasonable to label such entities as 'growing' or 'evolving' (warning: metaphors may be mixed during the course of this post).
After initially experimenting with the processing visualization environment I abandoned it in favour of 'good old' Flash (lens widget induced trauma now mostly repressed). Next up, choosing the basic metaphor (plane, terrain, tree, radar)? Hmm, let's try a variation on the hyperbolic tree (naturally, you say). What to represent? ... ah, let's go with pedigree, generation and mutation (replies which have different subject line to their precursor). A few hours later and presto! Coloured graphs everywhere.
A thread is born
Let's try out this visualization widget by using it to represent a popular recent thread (To the team of DPreview: Do a more serious test of 50d!!! - Canon EOS 50D - 10D forum) in its early phases (6 hours after OP).
Visualization of the To the team of DPreview: Do a more serious test of 50d!!! thread (6 hours).
Colour represents time. Connections represent lineage. Distance from OP (black dot) represents 'generation'. Dots outlined in black represent changes of subject.
So what are we seeing here? It all begins with the black dot representing the original post (OP). Sprawling out from the OP are several dead-end responses and one dominant response (A) which has sparked some conversation. Interestingly this 'successful' response was the first in the thread to change the subject line (to "DPreview has done retesting 50D... Look").
So, can we predict anything about the ongoing development of the thread by looking at the above image? Well it's a fair bet that conversation is going to follow on from A or one of its descendants, though it's also worth keeping an eye on the more recent responses to the OP too (B & C, also sporting altered titles). Let's take a look at how things have progressed 35 hours on.
Visualization of the To the team of DPreview: Do a more serious test of 50d!!! thread (41 hours).
Colour represents time. Connections represent lineage. Distance from OP (black dot) represents 'generation'. Dots outlined in black represent changes of subject.
Forty-one hours in we can see that the An branch (descending from A) did in fact continue a little while longer but has now run cold, instead branches Bn (descending from B) and Cn (descending from C) dominate the landscape. Note also that several variations to the OP have occurred since B & C, with some small success, suggesting that the OP's potential hasn't been fully exploited. This becomes more apparent as the thread progresses.
Visualization of the To the team of DPreview: Do a more serious test of 50d!!! thread (59 hours).
Colour represents time. Connections represent lineage. Distance from OP (black dot) represents 'generation'. Dots outlined in black represent changes of subject.
Forty-five hours into the thread a response (D) to the OP appeared (with the altered subject line: "Future reviews from dpreview will loose value"). The offshoots of D immediately starving previously dominant branches (Bn, Cn) and remained dominant until the 150 post limit was reached.
Time well spent?
Besides learning more about the flash 9 drawing API (note to self, read up on flash 10 drawing API) this mini-project has given me a better understanding of some of the nuances of how dpreview forum threads typically unfold. Of course these colourful diagrams don't tell the whole story or even a noteworthy fraction of it, but they do depict some latent trends which can be investigated via other means. For example I'm keen to investigate further the idea that responses to 'cold' threads garner more interest if they sport new subject lines versus the standard 'Re: ' prefix (another post perhaps).
More
I've included a few more diagrams of other threads below (click thumbnails for larger versions). I'd be interested to hear your thoughts about any more patterns you think might be worth investigating. Also, if you've got an idea for a visualization approach which may highlight other aspects of the forum (and yield some more eye-candy to boot) let me know.
Really nice plots! Great visualizations.
On the general subject of making sense of online conversations, you might check out the work done with Netscan by the folks at Microsoft Research.
Posted by: alan | Nov 28, 2008 2:53:54 PM
This is a really nice vis -- I hope you don't mind, I submitted it to the Infosthetics blog. Here in the InfoVis research community, I think we often miss some really cool work because we are too busy with our "academic" pursuits. I can't believe you pulled this together in a few hours, and it is so clearly useful... congrats! One criticism: there's a lot of work showing problems with rainbow colour scales in terms of perception of ordered relations, ability to discern differences, etc. You might want to look up Colin Ware's text book.
Posted by: Christopher Collins | Dec 2, 2008 7:29:34 AM
My first reaction to your post was Wow, this is fantastic! But then looking at the charts I can`t help think that they present no more information than just quickly scrolling down a forum topic. Sure, this is a lot more pretty with all the shapes and colors, but still, I think your superb (so far) efforts could be used to tease out some sort of information on the nature of threads.
For example, a long sub-thread with just two or three posters usually represents an argument. A long thread with no sub-threads and very few repeat posters usually represents a call to submit photos.
Also, because everyone follows a different approach to changing the post topic, any changes there are not very indicative I`m afraid. The time-line doesn`t say much about the thread either IMO.
Also, if a single post somewhere down a thread gets many replies, that post probably has a great importance in the thread and should be highlighted.
And also, some members post multiple times in the same thread (call them high contribution posters or something). Might be useful to highlight these with the dots outlined in black or gray or shapes (square, triangle).
Some suggestions:
- Posts with lots of replies should should be red (more than 5), and those with no replies should be blue, with other colors in between as per color scale, to indicate interest in the direction of discussion.
- Sub-threads with many posters should be thicker than sub-threads with only two posters to indicate popularity of discussion (thick) or repetitive/hair-splitting argument (thin).
- Repeat posts by the OP should be signaled with black dots in successive generations (so we guess at which sub-thread is still more or less on-topic).
- Anyone with more than five posts in one thread gets posts signaled with black/grey circles or squares or triangles.
Posted by: beuler | Dec 2, 2008 7:36:56 PM
"How and where to start? Well, there are myriad metaphors for the concept of 'conversations', but I like to frame them as Darwinian competitions between ideas in which the 'fittest' ideas 'breed' more 'successful' responses thus carrying on the line (of conversation ;)"
LOL you'd like to do that. Since you're intent on being scientific, why not try to define those concepts?
Then when all is said and done, run some code against the existing forums and see how well they rate.
Posted by: p | Dec 3, 2008 3:18:02 AM
...again, fix this so that a person can edit their replies...
"What to represent? ... ah, let's go with pedigree, generation and mutation (replies which have different subject line to their precursor)."
and you're not modeling the dpreview forums accurately, people here often change the subject line simply to avoid having to type text into the text box that someone has to actually read, to understand the nature of their reply.
Like
Re: why is the G10 so bad?
---it's not!
in general you are raising the interesting question of whether something is actually what you think it is, not to mention whether or not it means what you think it means. Or of course, whether it will become what you want it to become.
Posted by: p | Dec 3, 2008 3:24:24 AM
...ultimatey what I think is that all of this will most likely go for naught, as the forum moderators already know what they want to have and don't want to have, and you with your "scientific analysis" will just intrude upon their Grand Order.
What they basically have is a series of cozy little nests for their favorite posters to converse in, online. Perhaps a new poster might enter and rise to some significance in the pecking order, perhaps not. But certainly the most important issue will be how and why they challenge the existing pecking order.
You are dealing with photographers, not rational logicians.
What do you think they want to see on their forums? Whatever helps them to maintain their overblown egos. There used to be a popular saying, in that they would rather have a few great shots than a bunch of average shots. Why now concern yourself with the length of a thread? Supposedly it is not quantity that you seek here, but quality, yes? So every mouse will have a different opinion of what is and what is not quality and if you think that you are going to rationally win them over? Welcome to the "banned" list :)
Posted by: p | Dec 3, 2008 3:32:13 AM
Hey Jaysen,
well done. The graphs look great and does exhibit some interesting information, but I would rather visualize the time between the post as distance between nodes and use the color to show the mutation/generations. I would be very interested in the time wise
evolution of the discussion. But with the color scheme this is hard to get right now.
Another interesting visualization would the network of people involved in a forum and to identify groups. But I guess the forum moderators do have a sense of the groups already.
/lars
Posted by: Lars Kirchhoff | Dec 3, 2008 7:37:30 PM
If you can draw anything from your conclusions you should stop prefilling the subject line and still require it. I think it would improve the forum as it would give you an idea what's in the post.
Posted by: Android | Dec 4, 2008 9:28:41 AM
I'm loving the initiative here. DPReview's forums certainly need some TLC.
I'd be inclined to look to: a) allow post modifications (perhaps use a rule that posts can't be modified if responses are already present?), b) allow for self-moderation (the ability for posters to mod other posts down below user-selectable thresholds - like Slashdot - keeps trolls and fanboys at bay)
Posted by: Wireless Willie | Dec 4, 2008 4:39:51 PM
I think you should just print those visualizations in some large poster size and frame them- modern art...
might pull in some $$
BM
Posted by: bmadau | Dec 5, 2008 1:36:46 AM
Nice presentation but looks like this technique won't show correctly threads with multiple sub-threads on the same topic. For example in this thread I’ve suggested a remote testing technique to answer the OP question. http://forums.dpreview.com/forums/read.asp?forum=1009&message=23685973
And the results and other responses were in a few other sub threads.
I guess it would be more interesting to see large trends from the whole forum rather than from just a single thread.
1.
It would be interesting to see if there is a correlation between a total number of posts of a user ( and 2nd parameter - his registration date) To the number of replies to his threads and his posts. (Also number of views of his threads and posts.)
Being few years on dpreview forums sometimes I have an impression that people read the right column of the thread list first (i.e. the name of the poster).
2.
Many posts are being viewed. But only some get answered. It would be interesting to know the parameters of the "most answered post". Things like: Topic, length, links, images, poster reputation (posts, Registration date), day of week/time, maybe some keywords? etc...
3.
This brings me to the next question: Some posters write a single line and get the whole attention while others conduct tests, research and write long posts that sometimes get totally ignored. Is there any correlation between post length and number of its replies? (and their length ?)
4.
Search engine hits (internal and external). Which posts (parameters) get forgotten and which live long and serve future readers? Is there any correlation between the numbers of search engine hits on the post to things I’ve mentioned above? (This might be particularly interesting to dpreview as an information resource.)
5.
A lot more....
Posted by: madcat108 | Dec 5, 2008 2:05:59 AM
One thing to avoid (imho) would be to call mutli-response posts "successful". More often than not, the very reason a post gets a multitude of responses is because if it's lack of correctness. (and, as has been mentioned before, vice-versa as well)
Posted by: Wedding Joe | Dec 5, 2008 2:22:29 AM
"a) allow post modifications (perhaps use a rule that posts can't be modified if responses are already present?)"
...that's the problem with modifying posts, they can be completely changed and have no relation to the original post that people have replied to.
This is one good thing about not allowing people to edit posts at all. But then they have to issue another post to correct what they said and you get a bunch of pedants who reply to corret the original post. Or people who get confused by it.
But it just goes to show that if you can't even get this simple thing right and solve this simple problem that a forum as a whole is really an intractable mess. It's like anything. Allowing people to do it can cause as many problems as it solves and then you have millions of people who try to exploit it.
Posted by: p | Dec 5, 2008 8:00:51 PM
....one other thing that you will have trouble fixing, or defeating, on this forum: the self-anointed "expert", who *knows* the truth, who is just plain, flat wrong. But because he has X years shooting portraits/studio/action/landscape photography with Nikon/Canon/Pentax/Leica gear, he most-certainly knows what he's talking about. Except when it comes to logic.
Case in point (and I'm sure that there are many more)
http://forums.dpreview.com/forums/read.asp?forum=1032&message=30254678
Posted by: p | Dec 6, 2008 7:41:05 AM
...also this whole thing about "black spots" with the 5DMk2 that seem to be a function of exposure and the raw-converter of choice.
I can't say, I have yet to see them in any 5dMk2 file that I've converted with dcraw.
Yet I have yet to see anything about high-ISO streaking under lights at night that I see with the D700, so who knows. Are the fanboys that powerful? Was it just my camera? Who really knows.
Posted by: p | Dec 6, 2008 7:43:06 AM
Regarding editing posts : Probably a good compromise would be to allow to *Append* to the post instead of edit it. This way people can write updates or fix their mistakes, post updated links, etc...
Posted by: madcat108 | Dec 6, 2008 1:03:45 PM
"3.
This brings me to the next question: Some posters write a single line and get the whole attention while others conduct tests, research and write long posts that sometimes get totally ignored. Is there any correlation between post length and number of its replies? (and their length ?)"
...that's because posts beg a comment in type, as a reply. And of course, again you are dealing with photographers here, they are not likely to write well-researched, logical, factual posts in the first place. You are much more likely to see one "opinion-piece" after another. Some are short, some are long, that generate responses that are short, long & in-between.
The real question here is why does dpreview want to have forums in the first place and what do they want to see ont heir forums? Each user has to answer that question for themselves, as well. Is their goal to post on dpreview and to be widely-read on dpreview or is there some other goal? But given sufficient time all of these questions are answered by facts and made clear with hard data.
Posted by: p | Dec 6, 2008 9:51:03 PM
...and this one...
http://forums.dpreview.com/forums/read.asp?forum=1037&message=30243991
a good half of the posts are people replying just to tell the OP that they like his work.
Posted by: p | Dec 6, 2008 11:21:33 PM
plus when you think of it...
http://forums.dpreview.com/forums/read.asp?forum=1032&message=30260051
...this doesn't actually make any sense, technically. The point of the microlens is to increase the active area of the sensor (the fill-factor). It is not a prism *over* the microlens layer. The microlenses may act as prisms but the light is merely being filtered, collected and measured, at this point. It isn't going to get refracted into an adjacent RGB pixel. It would have to get refracted well away by the microlens layer, many pixels away, for image CA to be increased as a result of prisming in the microlens layer, and that's simply impossible...if the microlens layer induced that much diffraction then the sensor itself would be useless for off-normal imaging. This is all coming out of the lens itself, it is a product of increased resolution, not microlens-layer prisming. It is simply more visible at 100% viewing because of the higher resolution.
Anyway who is to say what is technically wrong or right here? There's no accounting for technical accuracy here in these forums, what you see here is a matter of democratic vote as much as anything else.
Posted by: p | Dec 6, 2008 11:33:47 PM
After a slow start in the comments, there does seem to be a little interest in this after all.
@Lars Kirchoff
Early in the development of this visualisation (about 2 hours in) I had a version which represented 'age' via distance from OP, but if anything it was more difficult to understand as posts would 'clump' together (i.e. almost on top of one another). Logarithmic distance was a little better, but i don't think anyone ever 'gets' logarithmic scales. Still, it's a good point and I still have the code so it may happen.
@bmadau
Thanks. Though I think the lack of irony and the possiblity of utility disqualifies this as 'modern art'. On reflection the algorithm makes poor use of space (lot's of 'dead' space), fails the 'squint test' and overall requires too much imagination on the part of the viewer. I have a few more ideas courtesy of the coral exhibit at the Natural History Museum, now all i need is some time.
@madcat108
Interesting ideas. With regard to the 'append' idea I think it has merit in the sense that it prevents revisionism and reduces gaming, but also hurts readibility. I'm leaning more towards the idea of comment revisions myself (example http://www.stackoverflow.com).
Posted by: Jaysen Marais | Dec 8, 2008 11:00:57 PM
Interesting post, brings out the nerd in me :-) is the graphical representation something you have thought up yourself (in which case you can call it a Jaysenogram) or where did the inspiration come from.
Posted by: JP | Dec 9, 2008 8:05:32 AM
This is just absolutely fascinating. A remarkable image.
www.burrellcolourimaging.com
Posted by: Burrell Colour Imaging | Dec 11, 2008 11:11:02 PM
Come on man! We're talking about a mere forum, but somehow you've managed to turn it into a scientific topic. hahaha
And what have we learned at the end of the day? Nothing 'noteworthy' according to your own conclusions.
Looks like Amazon still has plenty of money to waste..
Posted by: xaj | Dec 12, 2008 8:57:44 AM
Instead you should spend some time on developing a non confusing way to post comments here. A way that doesn't include two 'post' buttons and one 'continue' button on the same screen.
A way that doesn't say " Your comment could not be posted. Error type:
Your comment has been posted."
THAT would be a lot more usefull.
Posted by: xaj | Dec 12, 2008 9:03:06 AM
@JP
It's an original style as far as I know, but I'm sure there are similar diagram styles out there. As for the name, I think it needs work
@xaj
The only resource wasted in the production of this post was my personal time.
Posted by: Jaysen Marais | Dec 12, 2008 10:27:51 AM