SMTM

This is the java implementation of paper Multi-label Dataless Text Classification with Topic Modeling, Daochen Zha, Chenliang Li

Requirements

• Java 1.8
• Java Libraries (download lib.zip)

Data Preparation

Dataset directory should look like this: (The code uses both training and testing documents for learning but only computes metrics based on testing data)

├──dataset
   ├──train
      ├──[document 1]
      ├──[document 2]
      ├──...
   ├──test
      ├──[document 1]
      ├──[document 2]
      ├──...

categories file:

[train/test]/[document] [category 1] [category 2] ...

For example,

train/0001 C1 C2 C3
test/1001 C3 C4

says there is a training document 0001 in train/ with categories C1 C2 C3, and there is a testing document 1001 in test/ with categories C3 C4

seedwords file:

[category 1] [seedword 1] [seedword 2] ...
[category 2] [seedword 1] [seedword 2] ...
...

document:

[token 1] [token 2] ...

How to Run

The main entry is in src/main/BiasedGPU.java. Please specify some paths and the category number before running the code:

• catsFilePath: the path of the categories file.
• dataRootPath: the root directory of the dataset.
• seedwordPath: the path of the seedwords.
• catNum: the number of categories.