CSLI/CALO - GUI Usage HOWTO

This document contains instructions to developers for installing our CALO software and performing processing of a meeting using the debugging and development GUIs. It is intended for the SRI integration team to help get everything properly installed and debugged. See CaloApiHowto for information on how to accomplish these tasks programmatically using our service APIs in the deployable system.

Installation

1. Create a local configuration file
   1. Create a file calo/lib/config/calo.local.config with the following contents:
      • inherits=calo.config
      • calo.use.mercuryhome=false
      • opi.file.archive=PATH/nomosarchive PATH is the absolute path to directory where you want to put nomosarchive (e.g. D:/workspace)
      • corpora.orig.path=[PATH] PATH is the path to look for original unmodified corpora
      • file.lookup.path=[PATH] PATH is the path to look in for relative references to files in the annotations

Running

1. Start the Service Manager
   1. Start the Service Manager by Execute calo/run/Service Manager.bat or calo/run/Service Manager.sh
   2. Verify the CSLI Service Manager starts up
2. Import meeting data into CSLI KB framework
   1. Verify corpora.orig.path and file.lookup.path point to the directory contains all the meetings you want to import
   2. Run the ImportCaloY5Recordings importer from the KB Service GUI
      1. Select KBServiceImpl? tab from Service Manager
      2. Click "Run Importer" button from KBServiceImpl? tab. it will pop up a frame
         
         
         
      3. Select ImportCaloY5Recordings and click "Run" button
         
         
         
      4. This will re-import all data (both new and previously existing meetings)
3. Run/Test detection algorithms
   1. Generate initial action item hypotheses
      1. Choose the ActionItemIdentifierService/DecisionService tab in the Service Manager
      2. In the left "Available training data" panel, select the data you wish to train and hit "Train"
      3. In the central "Available meetings" panel, select the meeting for which you wish to detect and hit "Process" (skip previous step may result in trained all the data in the left panel)
         
         
         
   2. Generate a priori topic segmentation
      1. Choose the TopicSegmenterService tab
      2. The TopicSegmenterService GUI strictly follows the API methods. Please refer to the JavaDoc for more information about the meaning of the different arguments. Warning: Some of the methods can take up to one hour.
   3. Addressee...
      1. Coming soon...
   4. Parser...
      1. Coming soon...
4. Perform browser feedback
   1. Open up your meeting browser
   2. Open up the newly recorded meeting
   3. Review the results and supply user feedback
   4. Confirm your changes
      1. This action will save the feedback data to disk
      2. You can verify it has been created in the Action Item Service GUI
   5. Retrain the ActionItemIdentifierService:
      1. In the right-hand "feedback session" panel, choose the new feedback model and hit Interpret;
      2. This should produce a new training data set in the left-hand "training data" panel (hit Refresh there if you don't see it immediately);
      3. Hit Train to retrain on all available data (which will now include the new feedback). No need to select any training data files, it will use them all by default.