DKPro TC - Processing Modes

DKPro TC supports the following processing modes that correspond to typical classification setups in Natural Language Processing, which are shown in the table below:

  Single-Label Multi-Label Regression
Document Mode x x x
Unit Mode x x x
Sequence Mode x    

Document Mode: Classifes whole documents, classical use-case is for instance E-mail classification into spam or no spam (single-label)

Unit Mode: Sub-document classification, this is a special case of document mode where the classification focuses on two or more smaller portions of the text body, for instance classyfing the age-range on several user-comments under an article where both, the article and the user comments are hold together (single-label).

Sequence Mode: Classification mode for sequentially dependend information where the prediction of the preceding element is relevant for the prediction of the next one, for instance in part-of-speech tagging where the prediction sequence of word-labels carries meaning for the next word (single-label).

Classification using Multi-Label are for instance prediction of movie categories where a movie might belong to more than just a single category and the different categories have a certain correlation to (not) occur together. Regression is a special form of single-label classification where instead of a fixed label a numerica value is predicted.

The processing configuration is provided by two dimensions in a DKPro TC Experiment

  ParameterSpace pSpace = new ParameterSpace(
                Dimension.create(DIM_LEARNING_MODE, LM_SINGLE_LABEL), // single, multi, regression
                Dimension.create(DIM_FEATURE_MODE, FM_DOCUMENT), // document, unit, sequence
                .... //other dimensions
                );