Software for WSDL categorization

Installation

The source code and executables can be downloaded here.
The software is implemented in Java so you'll need a fairly recent Java installation (at least 1.5).
If you want to compile, you can preferably use Ant although it's of course also possible to compile manually.

Running

First, the executable ll-train from LibLinear must be included in the executable path. The executable can be found in the main directory but can be put anywhere on the system.
On Linux and Mac, the software can then be run directly using the script cv.sh. On Windows, you'll have to call Java directly or implement a simple BAT script that does the same thing as the shell script.
The script cv.sh contains experimental parameters such as the directory containing the categories (CAT_DIR).

Short notes about the source code

The source code is in the src directory.
The file WSDLReader.java contains the code to parse the XML and extract words. This can also be run on its own to print the words extracted from a single WSDL file: java -cp wsdl_categorizer.jar:utilities.jar:trove.jar wsdl_categorizer.WSDLReader FILE, where FILE is the WSDL file you want to parse.
The file WSDLCategorizer.java is where the experiments are carried out.

The output from an experiment

After some initial diagnostic messages, we can see the accuracy:
```
Micro accuracy: 14 / 14 = 1.0
```
Then we get the confusion matrix. Here the identifiers 0 and 1 represent the stock and weather identifiers, respectively:
```
0. stock
1. weather

0:  8 0
1:  0 6
```
After this, we have the per-class performance measures. The six numbers are the following: precision, recall, F-measure, number of correct guesses, number of guesses, and number of instances in the gold standard.
```
0: [1.0, 1.0, 1.0, 8.0, 8.0, 8.0]
1: [1.0, 1.0, 1.0, 6.0, 6.0, 6.0]
```

Then, after some additional diagnostic messages, comes the list of 10 most highly weighted features for every category:

*** For category "stock"
stock 0.1777026862733373
list 0.177284332878544
symbol 0.1716715433121654
currency 0.1590836172385301
value 0.1560681923697255
last 0.1552034102078856
quote 0.1548232007012907
security 0.1492285768311836
dollar 0.1460838390059707
all 0.1458992697230641
*** For category "weather"
weather	0.4236339104364145
zip 0.283701847546839
forecast 0.2218413016922719
code 0.2013819822112159
report 0.1723113339087071
city 0.166515341541315
day 0.1284451862296781
get 0.1155969040906293
u 0.1055154027396875
d 0.1026760476086771