Jan 21, 2016

Mapping Conversations on Twitter

I often participate in Twitter chats on topics I find interesting and useful.  It is via these Twitter chats that I often find new and interesting people to follow on one of my Twitter lists. I often check in on #innochat -- an ongoing conversation about innovation by those that practice it, or mentor others through the process of innovation.

Chats usually happen around a set of questions on the topic of the day.  The group facilitator asks the questions and the people participating answer the questions they please, beginning their tweet with an "A" and then the number of the question they are referring to (A1, A2, ...). Participants can answer questions, talk to the whole Group, RT each other, or have a side conversation with each other via @ messages.  

Figure 1 below is a network map of the conversations that happened on #innochat on January 14, 2016.  Individual participants in the conversation are shown by the blue circles, and the Group, as a whole, is shown by the square magenta-colored node.  The Answer node is shown by the green square.  A line/link is drawn from the person to the node s/he is directing their communication to.  Arrowheads show the direction of the flow and line thickness shows the volume of flow.

  • X --> Group: X is making a general tweet to the chat group using the hashtag #innochat
  • Y --> Z: Y is directing a comment to Z or RT'ing Z's tweet.
Figure 1
People who are just "listening in" or lurking are not shown on the map.  We can only map them if they "uncloak" and publicly participate.

Next we "zoom in" and look at only the conversations that were happening between participants in the chat.  We remove the generic GROUP and ANSWER nodes and the links to them.  We are left with just the #innochat participants who interacted with each other via an @message or RT.  That diagram is displayed in Figure 2 below.
Figure 2
Figure 2 also shows various node sizes.  The node sizes were determined by our social network algorithm that measures a person's involvement in the network of conversations.  Those more involved have a larger node.  Of course, the chat's facilitator, John W Lewis has the largest node.  Again, the links show direction of communication and volume.

For those that have participated in Twitter chats you know that they can be intense and that the hour goes by quickly as people respond to and discuss the questions posed.  It is nice to sit back afterwords and look at the social structure of the fast conversations that just whizzed by.

What key conversations are happening in your community or organization?  Who are the key players?  Who is missing?

Data Source: http://innochat.com/sites/default/files/transcript_of_innochat_2016.01.14.pdf


  1. I could see this being useful for some of the Twitter chats going on in Extension (e.g., #EdTechLN). Are you using InFlow for the graphing or something else?

  2. Thanks for the post, Valdis. How did you generate the raw data for the maps? We have a few Twitter chats I am aware of in Cooperative Extension. The one I participate in most regularly is the #edtechLN chat. It would be fascinating to see some of the network maps for that chat.

  3. Nice post about what is, who is, what isn't and who isn't involved. Making the invisible visible. That's what social technology is all about.

  4. This is fascinating. I love seeing how active I was and how much I was interacting with people. I, too, would love to know how to get diagrams like this.

  5. Thank you, Valdis, on behalf of #innochat, the innovation chat held each week on Thursdays at 12pm Eastern time. I'm impressed with the results of your analysis based on the data that we provide.
    These maps of our chat event on January 14th are an impressive demonstration of your network analysis expertise and facilities, showing the identity, relationships and involvement of participants in this conversation (about "New Knowledge" as the least reliable and predictable source of opportunities for innovation).
    As this chat was the first in a series of seven, maybe the whole series is an opportunity to extend your analysis.

  6. For the information of Stephen Judd, Bob Bertsch and anyone else who is interested, Valdis' maps are based on the Holosoft transcript of the #innochat chat of 2016 January 14. The transcript of each chat is attached to the weekly framing post so that the conversation can be reviewed after the event. For the chat of that week, on "New Knowledge", the framing post is here: http://innochat.com/innochats/date/2016-01-14/new-knowledge-unreliable-source-famous-innovations and the transcript (attached to it) is here: http://innochat.com/sites/default/files/transcript_of_innochat_2016.01.14.pdf
    These transcripts are provided to other chats, #smchat and #orgdna, as well as for larger events.

  7. John, Josh,
    Part of what we do at Orgnet, LLC is to teach/mentor others how to do social network analysis. We have software and 20+ years of knowledge/experience to share with clients.

    The PDF file was very hard to parse for the data we needed (PDFs are notorious for this problem, and the format is often chosen to obfuscate "public data") -- data analysts hate PDFs. If you want to see more networks, I will need the data in a CSV format, that will flow into our software easily. See if your #innochat transcript provider can make their data more accessible.

  8. Valdis,

    The #innochat transcripts are provided for offline reading by people interested in reviewing the conversation each week.
    These are not intended as a machine readable resource, so it is not surprising that it is not straightforward to scrape data from the PDF files.

    The transcripts are provided by me (on behalf of @holosoft). A couple of other regular Twitter chats subscribe to these transcripts, and we provide more complex navigable archives of conversations during longer periods to conferences and similar events.

    You and I have discussed the provision of information in Excel or CSV formats and I'm open for further discussion within the terms of service of the original source (Twitter, in this case).