Misc to start
We reviewed the current status of things, looked at the decision tree output done on CESR, looked at Mandala etc
We tried to use Hackey: http://tapor-dev.mcmaster.ca/~humviz/hackey/ . Worked fairly well but we also found bugs and improvements to add...
The goal for the weekend was to get an "A to Z" demo that used real data and worked through the interface, returning real results. When we left on Thursday afternoon, we were at "x."
Review of use case status and report (Sara and Catherine)
The status of the use case was reviewed and a report generated (see in the use case reports in the users and use cell page i.e. Sentimentality-status-April2008
Sara got interesting results so the UI team goal now it to create tools that would allow the same analysis thru the Monk interface in 1/10 of the time and effort 
Link to Mandala to provide a general collection browser and search (Matt)
- Matt will get help from Andrew. We already have the data (i.e. the Proxy call the get all the metadata of all the collection works with 3 attributes: collection, author, and date).
- Needs an export to save a workset.
- needs a Collection Search with Mandala new Toolset (or it can be a tool that replaces the current treebrowser tool)??
Status
Selections in the Mandala are now mirrored in the collection tree browser. Note that Firefox is the most reliable (and possibly only) browser to use Mandala within the workbench. A short video demonstrating how to select works from authors born between 1840 and 1890 is available here
Bugs/issues
- Sometimes a JavaScript error will occur when trying to load Mandala in the workbench. If this happens all you need to do is reload the frame in which Mandala is shown (the demo video shows how to do this).
Detailed review of workbench "search by example" needs (Andrew and Amit)
MOST IMPORTANT to get something that CAN be used by Sara
- Replace sending fake results to the workbench but send real results instead DONE
- Sending features back from the analysis (not just the ratings), ordered by most representative of one class to more represnetative of the other class (e.g. most sentimental at the top, and not sentimental at the bottom) STATUS: READY
- lA new component to display a table of features and their stats, with I hope some sorting function. Status: statted but there is a bug!
- linking features back to texti.e. when users select a feature, they can see which workparts of the workset contain it, and when the workpart is displayed the feature is highlighted Status: NOTHING YET
- Hookup all available SEASR flows : Status: MOSTLY DONE we now have 2 : 1) Decision Tree based on the training set and 2) Nayes Bayes Classification + Decision for everyting. In addition, the results return top N features, the median, mode, and average of features for both the workset and the classified results.
- linking in Amits' simple Prefuse tree browser applet to browse the feature decision tree of the training set, or the whole collection: Status L the tree data is coming out of proxy, but not in UI yet (should be by tomorrow)
- (and again: ) Linking between the applet and the workbench so users can see which workpart contains the features
- Import Sara's training data (Amit) Status: Sara has now cleaned up her data to be importable -
- To help the selection of the workset, a dynamic search in the titles would be very helpful so you could searcj for Tom, or Dickens and see all the work titles that include that string so you can build your workset faster. STATUS: Stefan is building a prototype for that. Not completely clear where it would go. A new search tool? part of the advanced search? Quick Search Tool To see it: goo to the workbench and start the project Stefan, and select the project toolset "Collection quick select"
An update (from Amit) on what still needs to be done (as of April 26):
- Amit has implemented the Flow Metadata call - it return the list of available flows and their description. Andrew
can now implement it in the analytics tool. It will display the list of flows available - a description and some other metadata like date
published etc (actually all of dublin core).
*Amit has hookedup the Itinerary that creates Decision tree for the training set with the monkmiddleware. Andrew can now
hookup the viz and we will have two working flows.
- Amit has imported Sara's data (both the workset and the training data) into the middleware under Sara Real project. Everything is fine except that
we don't have titles for these imported workparts in the rating table. In order for Andrew to be able to display the labels, Amit has created a
new call which retrieves the labels/titles for the testdata or traininglist in a workset; Andrew would need to hook that up.
*All of this has been moved to the production system on Friday the 25th.
Other less urgent needs discussed - that we should try NOT to address until the rest is done
The workbench needs to get descriptions of the Flows from the proxy and display them so that users can decide what to use -DONE
Providing a status line for debugging would really be helpful!
We discovered that the titles of the works at truncated! which is a serious problem. ACTION: Stefan send email around asking that the next version of the datastore.
About the 1-5 versus yes/no: From past experience we know that some users will want 1-5 while others will want yes/no. We also know that the data mining will work better with 2 classes than 5 for a training set of the same size. So what has been suggested and seems to be the best way to start in the short term is to allow 1-5 ratings, but users who want yes/no should only use 1 and 5 ratings. The next step is to allow users to specify that 1+2 should be considered as No and 4+5 as YES (or whatever mapping they want)... (In the future, the best solution would be to give users the choice of the number of classes, labels for the classes etc. )
https://apps.lis.uiuc.edu/wiki/download/attachments/2664073/CIMG3815.JPG
https://apps.lis.uiuc.edu/wiki/download/attachments/2664073/CIMG3816.JPG
https://apps.lis.uiuc.edu/wiki/download/attachments/2664073/CIMG3818.JPG
https://apps.lis.uiuc.edu/wiki/download/attachments/2664073/CIMG3822.JPG
https://apps.lis.uiuc.edu/wiki/download/attachments/2664073/CIMG3824.JPG
https://apps.lis.uiuc.edu/wiki/download/attachments/3738545/P1010249.jpg
