Very Simple Text Classifier Applets

These applets is some examples of a very simple text classification, e.g. for classifying spam, languages etc. It uses Weka as the machine learning engine. You will probably not use these applets for any serious text classification stuff. The main purpose of the system is just to play with the algorithms and maybe learn a thing or two about Weka and machine learning/data mining. See more about Weka here.

Canned examples

Here are some canned simple text classification examples:

If you, for some reasons, want to make you own canned examples, see the information below.

Documentation of the text classifier programs

Click on the Classify button to start classification. Show a different machine learning algorithm in the list Select algorithm. The Reset button clears the contents of all the fields. If you click the Classify button when all fields are empty, the canned example will be classified.

You can classify a new text by typing the text the field Classify new text:, and then click on the Classify button.

The two areas Text: and Class: contain the information about the texts to classify and its class (category etc). The class belongs to the text on the same row. You may add, delete or change any text or class. You can also add a new class just by typing the name of it besides a text row. If there are different number of text rows and class rows, the system will protest slightly; then just add or delete some rows or classes so there are the same number of rows.

Note: For implemental reasons the class "?" is added to the given classes. This class is used (internally) when classifying a new text.

In the result area a lot of information is shown:

ALGORITHM
the name of the machine learning algorithm used
DATA SET
hows the data set in ARFF format
A technical note: the program filters this data structure of just text strings and class attribute to a "Sparse" structure which is a "matrix" of word counts. This sparse structure is not shown here.
INFORMATION ABOUT THE CLASSIFIER AND EVALUATION
Here I mainly use Wekas wrappers for summary information:
- classifier.toString(): shows the classification model
- evaluation.toSummaryString: shows the evaluation summary and errors
- evaluation.toMatrixString(): confusion matrix. The matrix shows how the specific cases is classified and compared with the real class.
- evaluation.toClassDetailsString(): another view of classification errors. (E.g. "TP" is True Positive, "FP" is False Positive
- evaluation.toCumulativeMarginDistribution
CHECKING ALL THE INSTANCES
Here we compare the predicted values of all the instances with the correct classes. For each instance we show:
- probability distribution: Some classifiers outputs the probabilities that a specific instance is classified as a class.
- (The Weighted value is just some experimental thing.)
CLASSIFY NEW TEXT
If there is a text in the Classify new text: fields, this text is classified in the same manner as the training instances.

Parameters for canned examples

The canned examples just use the <applet>parameter tags (<param>) for populate the examples. The two parameters tag is param name= text + a line number for the text and the param name= class + a line number. The line numbers must start at 1, then increment by exactly 1 (i.e. use the natural numbers).

Here is an example:

<applet code="TextClassifierApplet.class" width="500" height="600">
<param name="text1" value="this is a class1 text">
<param name="class1" value="class1">
<param name="text2" value="this is also a class1 text ">
<param name="class2" value="class1">
<param name="text3" value="but here comes a class2'er">
<param name="class3" value="class2">
<param name="text4" value="this is for class2">
<param name="class4" value="class2">
No Java Browser
</applet>

Back to my Weka page
Back to my homepage
Created by Hakan Kjellerstrand (hakank)