Recently, Twitter user @soluzioni asked me the following question:
“Hi Christine! Since u seem to use Twitter lots, what are ur insights on the in-effectiveness of visualising threaded discussions? ..-)”
Although I love a good academic chat about Twitter, I didn’t understand this question at all. @soluzioni isn’t a native English speaker, and at times foreign users adopt too many buzzwords or odd habits. My first impression was, “What is ineffective about thinking of improving how we look at threading?” My second thought was, “Either threads are there or they aren’t.”
I ran this question and @soluzioni’s next one (“‘Nothing too clever’: a couple of connections / pivot points to who+what to facilitate context.” [-I think he’s trying to dumb it down for me]) past my rhetoric-busting husband. He suggested that perhaps @soluzioni, whose bio has references to mind mapping, was trying to create a way to represent visually the links between posts in a threaded discussion.
I can understand that. The “ineffective” part threw me off. Anyway, I’m going to work with the premise that @soluzioni is unhappy with the status quo of conversation mapping on the web.
Basically, a forum thread looks like this:
Main question
—>reply
—>reply
—->reply to a reply
—>reply
—->reply to a reply
—–>reply to the reply to a reply.
The first “main question” post is either at the top or the bottom of the page. Then replies are off-set under the main post. Replies to replies are offset even more. Many times, a user will click the “reply” button under the main question but continue on with a point brought up in one of the already existing replies. This means that if you want to follow the discussion, you have to spend the time viewing every single reply entry embedded under the “main question,” no matter what level of offset they are.
This type of chronological map is the simplest way to “follow” a conversation on-line. Unfortunately, the 140 character limit of Twitter makes it impossible to classify your conversation over multiple users and instances. At times people use hashtags (a # symbol followed by a term that reflects the conversation, e.g., #eaglesfootball) but hashtags still take up space and may not reflect all of the relevant topics in a conversation.
Some 3rd party Twitter API apps have attempted to thread Twitter conversations, and they do a decent job of capturing and offsetting immediate replies. But the more people you “have in your room” the more fragments will occur. Without hashtags (which people forget to add anyway), it’s impossible to capture all of the ideas that swarm during a Twitter conversation.
For businesses or people trying to gather ideas via social-media crowd sourcing, this is devastating. Even 1,000 Google alerts honed in on Twitter won’t bring you the little gem of an idea offered by a small-crowd Twitter user. I propose a different method.
I’d like to see a 3rd party app that can build keyword clouds from Twitter entries. I’d like this app to be highly customizable. I want to be able to mark the first “main topic” Twitter entry (a.k.a. “tweet”), then I’d like to modify the user span from which the app will draw keywords, then I’d like to put a time limit on it.
Here’s an example of using this fantasy application:
I put this entry into Twitter:
“Your company wants to start using Twitter as a business tool. How do you advise the use of social media to them?”
I then fire up my fantasy app. I link to my original entry.
I click the option of “all conversation” (as opposed to “only @replies”).
I put a time limit on it by clicking “in the next hour.”
The app first uses a Twitter search to see replies to me (@PurpleCar).
But, and this is where it gets innovative, the app also takes a catalog of all my followers and all THEIR followers, and makes a keyword cloud out of EVERYTHING they ALL tweet in the next hour. If one of your follower’s followers (3 tiers away from you) has more followers than say, 95% of all Twitter users, then that user’s tweets can be pulled into the keyword cloud (this would address “reach” — like a 6 degrees of separation concept — how far do your ideas reach around the Twitter community). The app, after an hour, would give you a reach score and a keyword cloud based on your choices. If your main topic’s keywords show growth (easier if they are very unique words), you can assume that you’ve started a conversation.
This keyword cloud along with some well planned Google blog search and keyword search over the next few days could lend a pretty reliable picture of what people in your community are talking about.
We are a LONG way away from this fantasy app. For the app to be effective, it would have to automatically abbreviate your words (e.g. “business” to “biz” or “bus”), have serious drive abilities (I have over 2,000 people in my room, most of whom are online when I am, and a lot of my followers have more than a few hundred followers themselves), and have total access to the Twitter API. None of those things is possible right now. It’s almost on the verge of artificial intelligence AND NASA-fast CPU’s AND server health and security at Twitter (the last of which is the most impossible option of all).
But, keyword clouds, twitter search and Google alerts exist, so with a bit of work you can fashion your own keyword cloud for a certain group of users over a certain amount of time. You’d have to work on a very small scale, but sometimes big ideas come from just one small voice. You just need to find a way to hear that tiny peep in the darkness.
I hope @soluzioni wasn’t too insulted, and I hope this answers his question about my insights into the threaded (or lack of threaded) conversations on Twitter.
UPDATE: 19 January 2009:
Erich from the comments reminded me via Twitter of tweetstats.com and wordle.net. I made a nice “wordle” of my Tweets – it’s a keyword cloud of the most used terms by me. You can see it here. It is for ALL of my Tweets since I started with the app in 2007.
Comments on this entry are closed.
I love the idea of the fantasy app, and I would use it; but it some ways, it doesn’t bother me because of what I use Twitter for. As an academic researcher, I use it for three things, no heavy duty analysis, rather:
– the instant quick hit in response to a question from me from one or two folk who know about something I don’t,
– general ‘sensing’ of whats going on up ahead in areas I might research in future
– random stuff that I’m not expecting, such as a recommended site for booklists.
So if threads were thgere, I might use it…but not too concerned, having fun as it is! (lwarren17)
Thanks Lorraine for checking in!
That is so funny, I use Twitter for almost the exact same reasons you do: crowd sourcing, trending, random fun (including good read recommendations!). Twitter can be so much more for so many businesses if only the way to harness the information was there. It’s becoming much more mainstream, and I can imagine a point where Twitter sampling will be more statistically sound than random mall survey culls. Online spaces have a way of attracting populations that are shy and reclusive that traditional sampling misses. Not that I want my fellow academic peeps longing for a better app, but the potential for study could be huge if we could just get over this data-mining issue.
Thanks again!
-PC
________________________________
I don’t know about anyone else, but a full fifty percent of the time I have tried to view hashtags the site did not cooperate. Got a message that told me to try again later.
So I gave up using it a as a tool.
Zed,
Hashtags are wrought with error, I agree. I just use tweetscan for them. The one thing hashtags have going for them is consistency in tagging. If a crowd agrees upon one certain tag, it is easier to follow a conversation using a search app like tweetscan or search.twitter.com.
Thanks for pointing that out, though, totally true.
-PC
I think BigAssTweet.com does something like this.
Jasper,
thanks, I’ll check out BigAssTweet.com. Definitely like the name! My husband and I met a Linguistics grad student at Penn who was studying the practice of tacking “-ass” on the ends of words, and how it changed or enhanced their meanings. “Dumb-ass” “Big-ass” stuff like that. I wish I could find her dissertation on that.
-PC
There are several apps that do exactly what you’ve described. Spiral16, Techrigy, Radian6, and CrimsonHexagon all come to mind. They are used for monitoring social media or brands. They all vary somewhat in their approach and technology. Each has pros and cons. All are worth exploring and there are others out there as well.
Dave,
Thanks! I’ll check out Spiral16, Techrigy, Radian6, and CrimsonHexagon for keyword clouds and conversation mapping in Twitter. I’ve also sent them along to my fellow podcasters at http://www.pushmyfollow.com . Maybe we can review them on a later show. I’ll be interested to see if they make the keyword clouds I’m talking about.
-PC
RE: [purplecar] Re: Conversation Mapping in Twitter: Keyword Clouds.
I will be interested in what you find out as well. They all make keyword clouds and have a process for displaying conversation links. I’m not sure if it is exactly what you are seeking. They are all fairly new companies and their products are in their infancy. It should get pretty wild as they all develop.
Thanks,
David
David Svet
http://www.spurcommunications.com
http://www.SPURspectives.com
I think Twitter is like speed-dating for conversations. I use Twitter b/c I like the speed-dating aspect: meet lots of smart people and hear all kinds of great ideas I never would have otherwise (like yours, which I was referred to by @soluzioni). So for me, I prefer the lack of complexity. When I want a fuller conversation, I lean towards email and blogs.
Pulling together a word tag of our followers comments, and their followers comments, probably could work out if you pushed the computation to end user; it’s a lot of bandwidth, but not out of the realm of question. The harder part would be putting identifying words and phrases with statistical changes in frequency over the period you monitor. Natural language processing has come a long way in the past decade, but it’s still far from pervasive.
My question is how could this information be used? If you knew you started a conversation, then what?
http://twitter.com/emorisse
Erich,
I agree. I prefer the lack of complexity for not only myself but for “regular”people (i.e. not social media geeks). I also agree that the whole design would be better if the enduser’s machines are used for the computation, but do you have any ideas on how that could work out?
The conversation mapping is of great interest to companies and marketing mavens. I’m sure if we build it, they will come in droves to use it. Think of the ego search factor alone! (ego search = googling yourself, etc.)
Thanks for commenting. Please let me know if you come up with anything.
-PC
Interesting topic. This is an issue not just for Twitter but for online conversation in general. We have technologies like FriendFeed, Disqus, Trackbacks, as on this blog, as well as other services that try to aggregate conversation threads into a comprehensive discourse.
Still, no one has got it quite right yet.
I would like to see better, and transparent use of metadata. Instead of using hashtags, what if Twitter (or any other commenting system) embedded a unique ID that would then be transmitted through, and associated with any reply or retweet?
That would make the job of your dream tool much easier, it could effectively do a Google search on that ID and then it would need to reassemble the discussion threads in some way.
— Ok — techie bit out of the way, what about visualising the conversation?
I think your tag cloud idea is a good way to visualise trends, and to illustrate how your meme has rippled through the outer degrees of your social graph.
However, this doesn’t address the issue of how do you go about following or joining this discussion in a sensible manner? At what point does it stop being a thread and start to branch into tangental conversations?
I don’t have the answer. If I did — I’d be looking for venture capital 🙂
Kai,
Thanks for jumping in. Good points. Let me think …
Hrmmm, unique ID’s. I can’t picture how that would work, how would the ID stay consistent among tweets if people are replying 3 levels down? Intriguing concept though. There must be some way, using the ID tag you mention and a search function to design the web to be able to have all related entries end up on one search results page. It would be acceptable if a few stray entries show up in the search results.
Keyword clouds don’t point you to a beginning or ending spot, so yes, it would be difficult to jump into a conversation just based on a keyword cloud. Good point. I wonder if we can have metadata behind a keyword cloud, maybe making each word in the cloud a link to the original instance of that keyword? What do you think, Kai?
-PC
Hello Christine, Kai and All,
I’ve also been invited by @soluzioni and I’m trying to bring a modest contribution.
Following the idea of an ID to the Tweet, I’ve added another ID for the sender.
So the second lines below are read: “user A started the thread. B, made his 1st reply to A’s 1st post…”.
A1: Initial Tweet
B1: Re A1
C1: Re A1
A2: Re B1
C2: Re A2
A3: Re C2
A4: Re B1
D1: Re C2
D1: Informational (to all her/his contacts)
A5: Re D1
E1(2 levels down): Informational
F1(3 levels down): Re E1
and so on..
To be able to follow in real-time with more ease the conversations, it could be interesting to separate Twitter’s home page in two (or at least give the choice for it):
– @replies.
– general messages.
So any reader could have the possibility to post an @reply or an informational message related to the initial topic he has been introduced by someone else. Sure there is no beginning or end in keyword clouds, but for the sake of clarity, the time scale would be the reference.
Well, let me know if that helps. I’m not sure yet how I can be useful on the visualisation of the clouds.
Ed
Edouard, bienvenido.
Thanks for coming!
Firstly let me say that you can use an application called Tweetdeck. That will “split” your Twitter homepage into columns that can be viewed all at once. I find this especially helpful because I tend to use Direct Messages in many of my conversations (we haven’t yet talked about how this practice causes impossibilities in conversation mapping — oy vey!)
Secondly let’s chat about your proposal. I get it, it makes sense, but I wonder how it would play out when thousands of people are chatting about one subject. I understand processors are pretty advanced nowadays, but the volume of conversation would be gigantic. Also, if processors could even handle it, the naming process would have to happen at Twitter’s servers, as simultaneous responses would have to be dealt with by one naming system. In other words, if there are unique IDs for each Tweet, then there must be only one server grid assigning the numbers. Perhaps there is a way around that, but we need to think in terms of data portability and speed.
The never-ending complexity of any numbering system makes the background metadata more complex. The legal industry would be the only one interested in the exacting minutiae of who-said-what-in-what-order. Naming (i.e. numbering) is not necessary for researchers like us to gather the trends in those conversations. This is why I proposed keyword clouds. I later proposed, based on comments, that perhaps each keyword should be clickable (linked) back to its first instance in the allotted time period. That will show us where it generated for the time span we are researching.
Simplicity is key. We want to design these conversation maps so “regular” people can follow them and gather data. We also want to keep scale in mind: some conversations may take place amoung 1,000 voices. I like your naming system for smaller conversations in Twitter but your system mimics forum threads. Forum threads are intuitive but can get quite cumbersome, to say the least. Perhaps we can use your naming system up to a certain critical mass. After that, the data automatically gets dumped into a keyword cloud.
What do you think?
-PC
Hi PC,
A slow and rapid response to tell You that I believe fast computers could handle the load of naming the conversations. But I’m not an expert. Yes indeed, nothing against keyword clouds which could work together with the ids.
Stay in touch!
Edouard