edgeryders/edgesense

Making sense of online conversations as networks

edgesense's Issues

Implement overlay information about metrics and bottom graphs

It would be nice to put some thinking into this, because clear explanations in the context of the network analysis – as opposed to referring the user to, say, a Wikipedia article – is probably helpful for takeup. For example, the whole button turns into its explanation upon mouseover.

Node count got stuck in Edgeryders

The node count is not updating. Has been stuck at 394 for a long time. Today, for example, we should have a new node, neo_collective, who wrote this post yesterday:

https://edgeryders.eu/visiting-unmonks

The post received 2 comments, so it should show up as a connected node to the graph, and the number of nodes should have gone up. I am pretty sure there are more new users.

Extract the network layout configuration in an separate file

this is to make it easier to change and make it possible to have different layout parameters for different networks

further possible addition: an in-page panel to adjust the parameters and re-run the force layout

Implement an alert for old data

The processed JSON data should never be more than 24 hours old. If they are, Edgesense should warn the user.

Remove loops from graph visualization and edge count

It is not clear that loops have meaning in a social network of comments. Removing them improves the clarity of the graph.

Put metrics in a stack on the right-hand page

This will leave more vertical space for the network.

Refactor python code packaging

refactor / package python code in a more proper way, see here
https://packaging.python.org/en/latest/distributing.html

Implement n. of nodes/n. of edges among the metrics

Instead of density and average distance.

Add Google Analytics code to the page

Including action monitoring

Rescale metrics time profiles

They should show the time profile of the whole period covered, highlighting the point of time the user is looking at.

Mis-titling of the license URL field

Adapt the layout to accommodate isolated nodes

In the Python script:

• run the Louvain algorithm as normal. After that is done, select all nodes with degree zero. Each of them is the only member of a class of the modularity-maximizing partition (shows up with its own color in the graph).
• Assign them all to the same partition class, and impose them a fixed color, for example white.
• Is there a way to influence the way colors are assigned to partition classes? It would make sense that the most numerous classes have primary colors, leaving the various beiges and shades for the classes with 2 or 3 nodes.

In the Javascript script:

• try increasing the "scaling" parameter in the Force Atlas algorithm by a factor 10. From your rendition, I assume Force Atlas does not touch disconnected nodes.

Implement centrality-based and degree-based color coding

Implement alternative color codings:

• high-degree nodes are on one extreme of a color spectrum, low-degree on the other.
• high-betweenness centrality nodes are on one extreme of a color spectrum, low-betweenness centrality on the other.

Add better metadata to the download functionality with DataPackage

As of now, the graph data are downloadable as GEXF. In order to make this open data, we need licensing. While we are at it, we add better metadata with OKFN's Data Package standard: http://data.okfn.org/doc/data-package.

Please note that it should be possible to configure Edgesense so that the data are made open and downloadable. I can take the metadata from the GEXF itself, save those on license and source.

Add the ability to specify the source jsons with an argument to the script

Are there node IDs?

(For the purpose of writing Tulip code on top of Edgesense data)

I need to define a function that finds nodes by ID. What do we use exactly to identify nodes? I see a field called "name" that looks like the best candidate functionally, but it may be a bit too open for research purposes. There is also a "networkx_key" field, but it seems set to zero all the time. I guess the question is: what does networkx_key do, and can I use it as user ID?

FYI: "name" stores user IDs in Innovatori PA (4023, 18952 etc.). In Edgeryders it stores actual usernames (John Doe).

Change metrics in the stack!

1. Number of nodes
2. Number of edges
3. Average distance
4. Modularity

Allow using remote json files

Design the tutorial/test

Create about five questions/answers that will make up the tutorial.

Enhance network layout in the case of disconnected communities

in the case of disconnected communitites the network layout makes the smaller components float away from the main component, it would be better to have them stick closer,

Rewrite bottom graphs in posts/comments per unit of time

Testing the tutorial

This may be a bug. I am doing the test on Matera, and step 3/5's answer at the time of writing should be "between 60 and 80%", it shows very clearly on the graph. And yet I get a 59% response. Is this still a drawing algorithm issue? Or does the algorithm select the last-but-one period, which is also compatible with the graph?

Show more clearly the user found in a search

When the search function is used on a user that has high degree (i.e. a moderator, for example), it might be daunting to visually localize her in her large highlighted ego network. (Try searching for hexayurt" on Edgeryders). A possible solution would be to show the balloon with the search user's name and data.

While we are at it, we could also add this to the double-click-on-a-node action.

Ambiguous answer in step 4 of the tutorial

On step 4 I found a case in which the share of comments written by non-moderator had decreased. This points to a decrease in sustainability, and I marked that answer. Upon pressing "next", I got:

The share of comments written by non-moderators has indeed in creased. However, if anything this means that the conversation might have become more self-sustaining lately, not less. Good work!

This recognizes my answer is right, however it phrases it like I had given the wrong one. The correct feedback is:

The share of comments written by non-moderators has indeed increased. _And yes, that indicates that the conversation might have become more self-sustaining, less dependant on moderators. Good work!_

I can't find the appropriate place in index.html to make the edit :-(

User search feature

To drive engagement, it might be useful to have a "vanity feature": you can look for individual users. That means people can look for themselves, and this might drive more folks to get familiar with Edgesense, and, more importantly, with networks themselves.

Recent (?) comments display normally in the graph, but not in the node's metrics

Check this out: user cenzopit has 2 links, both related to comments left today. You can see them correctly in the graph, but both in-and out-degree are set as zero.

Improve the display for larger networks

For example edge bundling:

http://www.win.tue.nl/~dholten/papers/forcebundles_eurovis.pdf

Already at the Edgeryders level (hundreds of nodes, thousands of edges) the visualization becomes chaotic.

Another possibility is adding a metagraph representation (see tulip: https://www.youtube.com/watch?v=hbhOfQgQbZY) , where the metanodes could be the classes of the modularity-maximizing partition, already computed by Edgesense. Problem: the Louvain algorithm is non-deterministic, so this representation is blurry at the edges.

fix window resizing bug

Time frame with end date + duration

Implement a way for the user to select a time window, excluding interactions that happened before and after the limits.

Beware: past interactions may still have a social influence in the present. The interpretation of the network is slightly different if the "start" slider is not set to zero.

Implement the possibility for the user to choose a "segment width", then drag-and-drop the segment (choosing end time and interval duration, rather than start time and end time.

Add live links to usernames

Links can be exported from the views.

New version of sigma too slow, the layout does not settle

Give it more time to compute, maybe?

Implement the interactive tutorial

Includes:

• a "start the tutorial" button
• the interface for submitting answers and receiving feedback
• a facility to store responses for surveying purposes.

Incomplete help text when configuring the Drupal module

Additionally, if the field "Tutorial Upload URL" is empty, the configuration page will report an error and refuse to save.

Temporarily, write
http://yourdomain.com/edgesense/tutorial_upload

Add info on number of users/comments

The dark blue frame with number of nodes could report:
Number of nodes: 112 of 426 users

And the green one with the number of edges:
Number of edges: 306 made of 2234 comments

The numbers of users and comments would be written smaller.

The information this is giving is: how many users are registered, but not part of the conversation; and to what extent the comments are "stacked" on top of each other in the same edge vs. fanning out.

"It's not too clear which the last period is"

Feedback on the tutorial from Matthias.

Fix the tutorial's text?

I would like to rephrase some things, correct at least one typo etc. Also, is it possible to use html tags (bold, essentially)?

Add inline help for interacting with the graph

Add a "?" icon next to the lock and filter icons to explain how to interact with the graph. Refer to the wiki page for help text:
https://github.com/Wikitalia/edgesense/wiki/Using-the-EdgeSense-demo

Write explanation for overlay information about metrics

Where shall I put them?

How can the share of comments/posts written by the community be greater than 1?

This happens in the MT2019 plot.

Zero-degree nodes?

The Matera network shows 2 unconnected nodes: Sara Errico and antoviggiani.

How can this happen?

Add an "export data" utility

• The data format should be easily reusable by most NA software packages (GEXF and GML are common to both Gephi and Tulip, and it seems NetworkX too).
• All data considered should be exported. Specifically, network data tend to be static, whereas Edgesense knows the time signature of user and comment creation, and these should be preserved. So, each comment should induce an edge, and the edge should encode the date of creation of that comment as a property. This means edges should be parallel.

Script the tutorial

Change the label of chart 3 from "Share of content" to "Share of community generated content"

The count of admin nodes seems wrong

When I look at the edgeryders.edgesense graph, I see about 10-12 black nodes, standing for administrators. But in the node count panel, if I tick off the "Administrators" box the count only goes down by 5. I thing the graph is right and the count is wrong.

In- and outdegree and edges drawn don't match

Look for example at user antoviggiani, to the northwest. She has indegree 2 and outdegree 4, but I only see four edges being drawn.

Implement overlay information about nodes

At least in-degree, out-degree, betweenness centrality.

Verify that a user writing one or more posts, but neither writing nor receiving comments, shows up as an isolated node in the graph

Ways to improve graph visualization without edge bundling

1. Remove loops as per Issue 44.
2. Change the layout to one based on arrows. This would eliminate approximately 20-25% of edges from the visualization.
3. Reduce to Simmelian backbone. Requires extra computatio, reduces edges by 60-80%.

Keep time profile of users?

What happens is a user is made a moderator, but had been active before being so?

Build an instructable that includes Tweet gathering and Edgesense installation/configuration

Users would have an installation of Edgesense monitoring a certain search string on Twitter. The string can be changed, but Edgesense will keep track of only one search at any given time.