From CS294-10 Visualization Sp11

Jump to: navigation, search


New Project

The slides are here: File:Bdon-viz-presentation.pdf

Oh, and here's the final poster: File:Bdon-viz-poster.pdf

Paper is here. File:Bdon-viz-paper.pdf Brandon Liu 18:38, 6 May 2011 (CDT)


Dan - Apr 11, 2011 03:45:11 pm

You definitely found a great topic, representing time! I think that hashtags and tweets would be an interesting data set because it is constantly changing, and also you are most likely to find overlapping attributes within tweets. It would be interesting how you could find overlaps in two different areas of different tags and optimizing the visualization display.

Dhawal - Apr 11, 2011 03:36:33 pm

Interesting visualization and looking forward to end-result. I would recommend working on one particular dataset since visualizations are closely linked with data. I'm sure you'll have clarity about data and your visualization technique once you start working.

Siamak Faridani - Apr 11, 2011 03:41:08 pm

I am incapable of interpreting streamgraphs properly but seems to be a very proper way of visualization for twitter domain and you have chosen an interesting domain. I am wondering if you can use methods like LSI and co-occurrence matrices to get a better neighborhood placement.

David Wong - Apr 11, 2011 05:09:09 pm

Interesting idea. Another data set that could be interesting are songs that are currently being played from users of music streaming sites, like hypem (http://hypem.com/#!/).

Matthew Can - Apr 11, 2011 06:23:39 pm

I like the idea of plotting overlapping relationships over time. Twitter messages are an ideal domain. Some other domains you could look at are word co-occurrences or Google search terms. I wouldn't worry about encountering sets that are geometrically infeasible. You can just restrict yourself to examples where your visualization works, and I think the results would still be interesting.

Sally Ahn - Apr 11, 2011 11:43:01 pm

I'm really interested in seeing what your visualization will look like. I agree with others that your domain choice seems very well suited for this type of visualization. Optimizing Venn diagrams for this type of data sounds like a pretty tough problem. The story told by this visualization may be hard to grasp, but it could nevertheless produce visually interesting images. Have you seen these visualizations of US population by county data? http://salavon.com/AmVar/AmVarStudy_Stills.php This uses 3D space, and I have no idea what story it is trying to tell, but the visualization is quite pretty!

Michael Cohen - Apr 12, 2011 12:25:39 am

One type of prior work that seems quite relevant to what you're after is the Sankey diagram. Here's one showing energy use in the U.S., and there are plenty of others on the web. In some ways they're akin to streamgraphs or general graph layout, but they (like your project) are often forced to deal with overlapping/crossing paths as gracefully as possible, so they might provide some inspiration on that aspect.

Krishna - Apr 12, 2011 12:23:09 pm

This is deep work :) I get the feeling that you are taking on streamgraphs which is nice - I really find it hard to interpret them. Have you thought of how you are going to compare your technique to streamgraphs, in terms of user studies ?

Saung Li - Apr 12, 2011 07:01:52 pm

It looks like there's some work on done on streamgraphs for Twitter: http://www.neoformix.com/2008/TwitterStreamGraphs.html You can check this out to see how this relates to your project. I personally don't like the readability of streamgraphs. It may help to add in more labels, especially for the axes.

Julian Limon - Apr 12, 2011 10:47:44 pm

Combining time series and Venn diagrams looks really interesting. I probably wasn't clear in class, but I believe that one way you could add more context to the visualization is by adding certain marks for milestones or events that trigger a change in behavior. For example, the time when a hashtag appears as a "trending topic" in Twitter's front page may be indicated with a line. This might be useful information who try to distinguish spam (those messages that only add hashtags because they are trending) from relevant tweets. Similarly, this could give companies metrics to track their social media activities.

Michael Hsueh - Apr 12, 2011 10:49:58 pm

One data domain I thought of that would benefit from this is global epidemiology. It's an area of great importance where people are always trying to find new tools and new ways of coalescing large amounts of data and aggressively trying to spot patterns or correlations at the earliest sign. I think tools in this direction have lots of data to crunch and are really quite consequential in terms of impact on lives.

[add comment]
Personal tools