ankit / see Goto Github PK

Ability to remove entity types, like "USA". This will reduce the number of false positives

In solving the Atlantic Storm dataset, I did not find the second view (following connected biclusters) helpful at all. There were a number of issues I ran into. The first view helped identify key items that were central to the plot, but exploring the connections was easier with the documents themselves. Here are some reasons.

1. The most related biclusters did not necessarily add new information. It might be part of the original social circle plus the word "FBI" which is just because of the documents. I seemed to be stuck revisiting the same information over and over again when following the tool. There was no sense of following a story, I could only do that with the documents. This is more related to biclusters as a tool than it is to our visualization as a tool.
2. In a related note, entities such as "FBI" which are of no practical relevance dominated the visualizations. Some way to eliminate entities as important connections would be helpful.
3. There was no way to explore in the 2nd view. You can select other nodes but that does not bring them to the center.
4. There was no good way in the second view to tell which entities belonged to which bicluster. They were just lumped together in one bicluster.

Color coding is used to represent entity types involved in a bicluster.

There should be a legend shown somewhere that tells you which entity types each color represents

Make the entities in the bicluster table view clickable. On click, only the biclusters containing that entity should be visible

Add ability to bookmark/save biclusters for later reference.

The user should be able to drag + drop nodes into a physical bucket space on the screen. Or a simpler implementation will be to right click on a node and add it to a bucket.

If the noise reduction features are implemented, we would want to export and save the modified data so that we don't have to redo all the noise reduction steps again.

A History view that shows all the biclusters the user has explored in a session.

Maybe useful for the user to easily go back to a previously interesting biclusters.

Will help discover important entities that are common across multiple biclusters.

Again, this contributes to reducing the noise in the data set