IBM Tool for Interactive Text Classification and Labeling

Overview

Requirements

Download

FAQs

Forum

Reviews

Platform requirements

Operating systems: Windows® XP (tested)

Software:

• Unstructured Information Management Architecture (UIMA) SDK (available here at alphaWorks®)
• Xerces Java™ Parser (tested with version 2.9)
• JDK 1.4 or above
• Tomcat (or other Web server).

Installation instructions

The bundled example text classification data set in this package is the Reuters-21578 text classification benchmark data set, hereafter refered to as reuters.

1. Unzip the zip file under some directory, such as C:\TICL. The following steps assume you have unzipped the file in C:\TICL. If you unzip to a different location, please change following paths accordingly.
2. Review the directory structure: Your C:\TICL directory will contain the following five subdirectories: configs, data, lib, new, and scripts.
   • C:\TICL\configs: This subdirectory contains all the configuration files under the directory reuters. For each new classification task, create a directory and put all the required configuration files under the same.
   • C:\TICL\data: Inside data, there is a directory reuters that contains training and test data for this classification task. Both traindata and testdata contain ten directories named after the topics; these directories contain actual files. These directories also contain a file classLabels.txt that contains a listing of the class labels. For any new classification task, that structure must be followed.
   • C:\TICL\lib: This subdirectory contains required JAR files.
   • C:\TICL\new: This subdirectory contains the files to be deployed in the Web server.
   • C:\TICL\scripts: This subdirectory contains the scripts for running different modules of the system.
3. Download and configure dependent software, as follows:
   1. Make sure that Java JDK 1.4 or above is in PATH.
   2. Download UIMA and copy the following JAR files from UIMA\lib to C:\TICL\lib:
      • juru.jar siapi.jar
      • uima_adapter_soap.jar
      • uima_adapter_vinci.jar
      • uima_core.jar
      • uima_cpe.jar
      • uima_examples.jar
      • uima_jcas_builtin_types.jar
      • uima_search.jar
      • uima_tools.jar
   3. Download and unpack Xerces (we have tested with Version 2.9). Copy xercesImpl.jar and xml-apis.jar to C:\TICL\lib.
   4. Installing and configuring Tomcat (or other Web server):
      1. Download and install Tomcat 5.5 (or any other Web server). We assume Tomcat is installed in C:\Tomcat. Again, please change the following paths as appropriate.
      2. Copy C:\TICL\lib\ii.jar to the lib folder of the Web server. In our case, it will be C:\Tomcat\common\lib.
      3. Create a file called root.txt in C:\Tomcat. Type the location of the data directory (in our case, C:\TICL\data) in root.txt; then save and close the file.
      4. Copy the directory C:\TICL\new to C:\Tomcat\webapps\new.
      5. Restart the Web server.
4. Run initializing test scripts from command line, as follows:
   1. Run C:\TICL\scripts\train.bat. This script will create ten files (test0.txt to test9.txt) in C:\TICL (as the data set is a ten-class data set); a directory xcas_output in C:\TICL\configs; a file dictionary.bin; and protrain in C:\TICL\data\reuters. The script will create a directory models in C:\TICL\data\reuters containing 20 files (features.selected0 to features.selected9 and nbm.mdl0 to nbm.mdl9). The script will create ten files (platt0.txt to platt9.txt) in C:\TICL\data\reuters\models.
   2. Run C:\TICL\scripts\test.bat. There are many test files, so it might take a while to complete. (It took 90 seconds in a 2.13-GHz machine with 1.5 GB RAM). This script will create two files: profact and lists in C:\TICL\data\reuters.

For usage instructions, please see the included readme.txt file.