Discussions > The words of TEL research, a picture from the TeLearn Open Archive
The words of TEL research, a picture from the TeLearn Open Archive
|
Nicolas Balacheff
191 days ago |
We continue our exploration of the universe of terms and expressions used in the TEL research area. You will find [ here] a first vision of what we get when exploring the resources available in the TEL OA. You may have some comments. This graph is not easy to analyse, a complementary tool may help, you will find it [ there]. Clicking on a word/vertex allows you to get a picture of its neighbours.
All comments and suggestions are welcome.
|
|
Thomas Ullmann
191 days ago |
Hello Nicolas,
quite interesting work. I would like to explore it a bit more. Is it possible to get the original .gephi file?
Thomas
|
|
Nicolas Balacheff
191 days ago |
That's in your mailbox, thanks to Emilie efficiency. We look forward to reading your analysis, comments, suggestions.
|
|
Thomas Ullmann
190 days ago |
Thanks Nicolas and Emilie,
I had a look and to questions came up. 1.) Why are the nodes directed? and 2.) Is the node colouring based on the partition of the network (as in community detection of social graphs. i.e. dense connection within the community and sparse connection between communities)?
|
|
Emilie Manon
188 days ago |
Hello Thomas,
The graph is a directed graph as relations between nodes are not necessarily symetric. "Open learning" is linked to "learners" but "learners" istself is linked to other words (arrows in http://maps.telearn.org/GexfWalker/ demonstrate that).
The node colouring is based on clusters. Nodes inherit their colours from the bigger nodes they are linked to (colours have been randomly chosen). This was the best way to identify different thematic clusters in our corpus.
I'm new to Gephi software and I don't know how "community detection" work but I'll be glad to learn more if you think it might be a better way to deal with our graph.
|
|
Thomas Ullmann
188 days ago |
Hello Emilie,
thank you for answering my questions. I had some time today, to apply the community detection idea to your data.
In the first picture, you see your data, but I made the graph a bit more compact by filtering the "long tail" of words based on the degree of each node. The colours are based on your clustering.
In a next step, I applied a community detection algorithm. A community shows a dense connection within the community and sparse connection between communities. Each community has its own colour. The size of the nodes is still based on the degree of the node.
In a last step I changed the ranking for the node size from the node degree to the community degree of each node.
Just looking at the last graph, the connected words make some sense for me, and some topics are apparent. What do you think?
|
|
Emilie Manon
188 days ago |
Thank you for taking the time to do all that !
I agree, community detection is a much more efficient way !The fact that I can easily see how many communities are detected, their respective proportion (%), words corresponding to each community, will be big help !
So I tried this morning and I've got 2 questions :
- i don't really get what you mean by community degree ? and does it correspond to the clustering coefficient ?
- spatialisation has changed from pic 2 to pic 3, is it caused by node size reattribution ?
For sure, I will reedit the graph with your method !
thanks
|
|
Thomas Ullmann
187 days ago |
This is quite interesting work, so I had to play with the data :-). I was surprised that with the community detection something somehow meaningful appeared.
With community degree I mean that instead of using the degree (in/out-degree) for the size of the nodes, I used the weights of the nodes in each community. This is the difference between the figure two and three and is also the reason for the different layout of the graph.
One thing that still puzzles me is that your graph is directed. In your example you say that "Open learning" is linked to "learners" but "learners" itself is linked to other words. To generate the data, do you use then some POS tagging of your sentences to detect some directed relations for example open learning "takes place in" learners?
|
|
Emilie Manon
187 days ago |
What I mean is that "open-learning" is linked to "learners" because "learners" is one of the top words appearing in "open-learning" close context (close context: 50 words before-50 words after the word). But "learners" is not directly linked to "open-learning" because other words appears much many times that "open-learning" actually do in "learners" close context.
Hope that helps.
|
