Masters of Networks: Ethnography

What do you get when you cross ethnography with network science? This is what we want to find out.

In this hands-on session, we  work with data from the opencare Edgeryders conversation, that has undergone ethnographic coding. Our goal is to find a rigorous, intuitive and beautiful way to synthesize hundreds of thousand of words in a conversation involving 200 people and over 2,000 ethnographic annotations. We will do this by building, visualizing, exploring and discussing various types of graphs. More inforrmation on preliminary work is here.

Masters of Networks: Ethnography is hosted by opencare and open to all. The research teams of Edgeryders and the University of Bordeaux will play host, providing data and skills for ethnographic analysis, data crunching, network visualization and software development.

Everyone is welcome. We particularly recommend the session to network entusiasts, ethnographers and anthropologists. We are not building a marketing instrument: if your only interest is in surveillance of consumers, this is probably not for you.

We will protoype graph visualization using Tulip: Data Visualization Software | Tulip

UPDATE - We also made a Facebook event.

Date: 2016-11-28 09:30:00 - 2016-11-28 18:00:00, Europe/Brussels Time.

1 Like

Edgesense II

Hey, I love these Masters of Networks sessions!

The whole Bordeaux data crunching network visualization team will join (Bruno, Jason and myself).

The next Masters of Networks session is a good time to announce our beta prototype of what could become Edgesense II. Go have a look, send feedbacl, as questions, criticize or like:

Explore the different views, zoom and click to access comments and posts:

  • Global view let you see all the data we have as a big chuny network - not that interesting after all, it's meant more for debugging
  • Tag view - this is one AMelia and Alberto want to play with and on which we expect feedback.
  • DOI view lets you look at a "local", smaller, view of the network. A tag-orinted DOI view is on the way and should be ready for the meeting if not before. @Jason_Vallet is working on it.

We also will provide the latest data dumps here. @Jason_Vallet will soon provide a link for you to download the data.

1 Like

I had to try the link a couple of times

but it eventually worked. Looks very interesting! After playing with it for 10 minutes I wonder what kind of navigation/search functionality could best be included.

Thanks for the feedback

The server hosting this service is only meant as a demo, we use it to experiment with different views on the data and collect feedback from users (community managers like @Alberto  and @Noemi, and ehtnographers like @Amelia).

The functionality strongly depends on the taask you are conducting. For example, Alberto is trying to understand how topics co-occur within discussions (not necessarily a same thread) so as to form emerging ideas/issues/… We designed a view on tags to help Amelia gain a high level view on how tags distribute across posts/comments, so she can reflect on her own activity.

You may have some other idea of a task/usage in mind, please share your thoughts.


1 Like

My interest would be

a couple of things.

One is high level community management - e.g. how to teach community managers to engage with people more effectively.

But I am also interested in the general phenomenon as I’ve worked with with percolation and networks a bit. There are a lot of aspect of which I am curious how they map onto a human network.

For feedback perhaps something like this would be helpful?

Getting ready for the workshop

In prevision of the Master of Network event, we have prepared a snapshot of the current opencare network we plan to take a look at during the hands-on session.

For easy (and fast) prototyping and exploration, we will use Tulip to visualise the data, so if you plan to attend and play with us during the event, try to install Tulip beforehand (freely available to all at

The Tulip file containing the whole network is available here. We precomputed three different “representations” of the network:

  • a global view of the forum (named Forum Network)
  • a user-to-user interaction network (Interaction Network)
  • and a tag-to-tag representation connecting tags which annotate common posts and comments (Tag Tag Network)

We also included in the file a Python script extracting the conversations annotated by one or several selected tags (showing posts, comments and the posting users).

The network has been created with data available today (24/11/16) but we will replace the file with an up-to-date version right before starting the session. For the truly brave and adventurous ones among you, an archive with the raw (anonymised) json files used to build the networks is also available here.

1 Like

Amazing work!

Thanks, @Jason_Vallet . You guys rock!

small steps on the qualitative side

Hi there @Alberto, @Noemi, @melancon, @Amelia, @bpinaud , @Jason_Vallet,@Amelia and I had a short meeting online to discuss some insights after November 28th session.  The goal for me is still to focus and understand if and where there is knowledge creation by “reading” representations (visual and textual) of OP3NCARE links and nodes. I’ll try to develop a primer (a index with some parameters, or a conceptual diagram …depends on the first findings) on what knowledge creation is and what processes can be understood in the paths we are sketching by network analysis…and share it with you for improvements and critics. On the opinion of @Amelia, we d’better tune and negotiate the research work between the richness and complexity of conversations and good synthesis.

Now, on a concrete level, here is a viable hypothesis and what i’m working on. I’m familiarizing with Tulip (@bpinaud probably i will need your magic) and sampling on few examples splitting subgraphs representing posts and subgraphs representing comments. My idea is to compare the two types of contents and see what to concentrate on to look for some findings and understandings.

That’s all folks

more news coming in next days…

Great work!

It’s lovely to see ethnographers like yourself and @Amelia embrace network science!

So many thanks for this work, we really appreciate it.

On another note: could you and @bpinaud do the following two things:

  1. For Bruno: go here and follow the instructions. We need the code pushed to the repo AND one nice picture. 
  2. For Federico: write a couple of paragraph explaining what you are looking for and if you found it in the data!

Federico, I have to say that your “deep dive” approach of following the threads connecting the most used codes was completely new to me. It’s really fresh thinking! I have been moving in the opposite direction, that of using the graph to draw a synthesis.


I will check later. It’s probably ok. Thanks!

What git repo

@bpinaud What git repo did you push the file and pictures to?


The one created by @Alberto: GitHub - bpinaud/Masters-of-Networks-5: Ethnographic data hackathon to design the opencare dashboard

Accept the invitation…

@melancon you have an invitation to the opencare Github organisation (since HH Bordeaux). Everything follows from that. :slight_smile:

Three points about a hypothesis i’m working on

So here is my reader’s digest for today with some detailed thoughts about the construction of my analytical model.

It is addressed mainly to @Alberto, @Noemi, @Amelia and @bpinaud

Jump to modelling in point 3 if you are not interested in some boring sociological speculations (point 1) and my first hypothesis (arguable and temporaneous…of course) (point 2):

  1. A strong starting point about “collective intelligence” consists in a statement of Emile Durkheim. I like to move on from classics! Lets’ see what he says…

As far as i recall: collective intelligence (society) trascends individuals by space and time and is detectable by collective representations visible in enacted practices such as performances, rituals, taboos, celebrations (mostly in 1893 and 1912 writings, i should check anyway). Durkheim had the idea of flows of ideas and values living inside people and making them synchronised the ones to the others thanks to society representing itself in them. …a sort of nightmare…but that was during positivism…and these were the first cultural theories… anyway what is sure (and many failed in interpreting his thought) is he never said that people part of a society think the same things and do the same things…he was for a organic view of society.

Of course here we are not dealing with holy/profane and other stuff from the aboriginal cultures from the australian outback studied by Durkheim, but the point stays in the relation between a community and its enacted practices (material as much as discursive). How a community represents itself in what it does?.. the recall of Durkheim in the theory of “communities of practice” (Wenger, 1988) is strong; infact Wenger sees the domain of specialised knowledge the way a community identifies itself by putting such knowledge in practice (an example for all: the physicians).

So, it’s about community, knowledge and what can be detected in terms of accountability of something textual (described): the practice, what people do and tell, or don’t do and don’t tell…(i’m still curious about lurkers…)

Now there are some pros and cons in our case, as the more I deep in the conversations and development of the OP3NCARE big narration (we are all making the history of it…) I remark a strong involvement of anyone, but what I also see is the need to keep separate the technical and the social as if they were two different worlds. In the process (ans if we want to tell about how the process is happening) there are time to time needs to move further steps in improving the technical side (let’s say the environment where the community lives) as the community changes or grows. Other times negotiations or silence happen…and there’s nothing more socially negotiated than silence (how much time do I need to stay silence to let you understand the lesson is beginning? 10 seconds, 12, one minute…? I used often as example/experiment to start my lectures). So I think about stories with many replies and a lot of effervescence and interest going on compared to stories with no replies, or just one or two…

Surely, surfing through OP3NCARE, we have a evident activity and all the stories written by people commenting and coming to know each others build in sociological terms a phenomenon. We must consider that these participants have an interest in posting their stories, have certain digital skills and be open care oriented we might. Probably one on ten of stories eligible for OP3NCARE are being posted, because information was not available, or there are no digital skills enough to post the story, or as institutional projects are not allowed to publish anything on the web without authoriarion, who knows…

Although we shouldn’t also forget the input and boosting of activities run by edgeryders staff, who is posting stories, commenting and connecting people. Now the point of connecting people is very important as it goes in the direction of shaping a interacted behaviour and, yes, the foregrounds of a culture based on sharing stories and reading around. That’s what I find be as much an elementary as an important point.

Said all this, lets’ move to construct the model…i’m still working to consider more authors and references (networked localities, virtual communities behaviour, etc…collective/connective…) to be included and what above is just some of the ideas i’m having…don’t expect too much methodology yet.

  1. We can get some elements for the dimensions of analysis considering:

a) what and where are collective/social facts in OP3NCARE

b) what can be considered a collective intelligent action or effect in OP3NCARE (this to detect and observe phenomena referable to the “community in the making”, i.e. the behind the scenes in terms of negotiations, contacts, tuning needed to get to a meaningful picture of what we see on the screen: each opencare story…i guess)

c) how much OP3NCARE categories and forms of classifications can be considered socially driven in their causes and effects, as infrastructuring the culture

All these elements could become parts of a model that could be used to perform some tests (see point 3) and see what happens when we add some details to check if the model satisfies the description of the processes with the data we have.

I still have not clear in my mind what the whole OP3NCARE networkscape looks like on a hand, while on the other -althought I understand some visualization and processing properties- Tulip needs much more practice for me to understand how to run and be confident with it. That is my fault of course.

What is sure is that we must understand not the single parts as parts, but as effects of a whole that is the ecosystem itself where identities, relations and objects become real and tangible…it is a matter of boundaries to work on, so lets’ check what these boundaries are, starting from what the OP3NCARE mouth tells.

  1. Now -on a peculiar level- all this bring me to compare what happens at level of co-occurences.

I’m interested in much frequent co-occurences as these are the semantic frames more redundant.

By such redundancy probably we can reconstruct and give meaning to how large discursive and thought processes ( I use a metaphor of human brain for OP3NCARE) start (the posts) and develop (the comments). To hold the meaning inside the network and see it emerging as bootstrapping process I would like to compare graphs made up only of posts to graphs made up only of comments.

From @Amelia I would need to know which is the sector of the network already charted by codes (can you provide a file to me or Bruno?)

From Bruno @bpinaud I would need three files (the coded sector network to get the graph; the graph of posts alone; the graph of comments alone) or some help to split graphs containing posts only and graphs containing comments only (still a newbie with Tulip…trying hard anyway).

Then I can read the codes and the related contents to have the concrete material to work on.

A next step will consists in re-reading the most significative stories tagged by graphs results and try to use a word tree (for instance to compare each story content to posts graphs and comment graphs results containing the bunch of stories picked up by TULIP processing the co-occurences.

Is this all too senseless to you?

Still working on to make the sketched model out of the hypothesis much keen and senseful (to me too)…

1 Like


@federico_monaco I read your comment a couple of times but I struggle to understand it. As soon as I do, I will write a reply .

1 Like

just consider the third part then

I was afraid it would have been unclear, but i thought was worth to publish my thoughts about while i’m still working on the conceptual model to use @Amelia 's work by the process of network analysis to mean possibly collective intelligence enacted practices.

Once done some evaluations and insights of the conversations i will come back to you.

Just read part three…

Ok, here we go

I am not sure I follow the part about collective intelligence. What we mean by collective intellligence in this context is:

opencare knows things that no one person in opencare knows

For example, opencare “knows” that the most commonly associated concepts are these:

No one “in” opencare can know this, because knowing it means accessing all the associations in everyone’s writing and normalizing them as a professional ethnographer does. Also, the associations themselves emerge in the context of an open conversation. So, even if you made one association yourself, you might not have made it, had you been alone (or in a different context).

FIles: over to Bruno.

1 Like

the ghost in the network


The biggest problem in analysing a textuality expressed by a collective (a collective representation) -for example a symbol, a totem, a taboo, a ritual, a practice- is how to understand it even if we are not part of it. This is the biggest problem for ethnographers and cultural anthropologists.

The hypothesis of what we should have here in OP3NCARE, by the network analysis done on coding, is a whole more than the sum of the parts that transcends individual posts and comments.

My goal now is to put in relations parts of the collective produced by the network analysis and compare them to the parts to see if concretely there is a correspondence between the representation we get and the practice of posting and commenting.

All this is so fascinating ( for me at least).

Useful work!

Well, you have my thanks. This kind of validation is useful work.

1 Like

files to produce

Hi all,

About the three files mentionned, I do not understand what do you mean by the coded sector network to get the graph and for the two others, what are the links between the nodes? It is easy to extract graphs with posts alone or comments alone (“equal value” clustering plugin may do the trick), but do you want a link between nodes?