OpenNLPDocumentCategorizerTrain

Synopsis

Train a document categorizer using the OpenNLP library.

This module is experimental.

Description

OpenNLPDocumentCategorizerTrain trains a document categorizer using the OpenNLP library. The documents and their class are specified by documents and categoryFeature . The classifier algorithm uses the document content specified by tokens and form .

By default the features are BOW but can be deactivated with bagOfWords . Additionally nGrams can be set to add n-gram features.

The classifier is stored in model . This file can be used by OpenNLPDocumentCategorizer .

Snippet

📋

<opennlpdocumentcategorizertrain class="OpenNLPDocumentCategorizerTrain">
    <categoryFeature></categoryFeature>
    <language></language>
    <model></model>
</opennlpdocumentcategorizertrain>

Mandatory parameters

categoryFeature

Feature where the category is read.

language

Language of the documents (ISO 639-1 two-letter code).

model

File where to store the classifier.

Optional parameters

classWeights

Weight of samples of each class. This parameter is useful to compensate unbalanced training sets. The default weight is 1.

nGrams

Maximum size of n-gram features (minimum is 2). If not set, then do not use n-gram features.

algorithm

Categorization algorithm. Must be one of:

• naive-bayes , nb
• generalized-iterative-scaling , gis
• perceptron
• quasi-newton , qn , l-bfgs , lbfgs , bfgs

bagOfWords

Either to generate single-word features.

documents

Elements to classify. This expression is evaluated from the corpus.

form

Form of the token. This expression is evaluated as a string from the token.

iterations

Number of learning iterations.

tokens

Tokens of the elements to classify. This expression is evaluated as a list of elements from the element to classify.

Deprecated parameters