Use opennlp to customize named entities



This article mainly studies how to use opennlp to customize named entities, label training and model application.




Training model

        // train the name finder
        String typedEntities = "<START:organization> NATO <END>\n" +
                "<START:location> United States <END>\n" +
                "<START:organization> NATO Parliamentary Assembly <END>\n" +
                "<START:location> Edinburgh <END>\n" +
                "<START:location> Britain <END>\n" +
                "<START:person> Anders Fogh Rasmussen <END>\n" +
                "<START:location> U . S . <END>\n" +
                "<START:person> Barack Obama <END>\n" +
                "<START:location> Afghanistan <END>\n" +
                "<START:person> Rasmussen <END>\n" +
                "<START:location> Afghanistan <END>\n" +
                "<START:date> 2010 <END>";
        ObjectStream<NameSample> sampleStream = new NameSampleDataStream(
                new PlainTextByLineStream(new MockInputStreamFactory(typedEntities), "UTF-8"));

        TrainingParameters params = new TrainingParameters();
        params.put(TrainingParameters.ALGORITHM_PARAM, "MAXENT");
        params.put(TrainingParameters.ITERATIONS_PARAM, 70);
        params.put(TrainingParameters.CUTOFF_PARAM, 1);

        TokenNameFinderModel nameFinderModel = NameFinderME.train("eng", null, sampleStream,
                params, TokenNameFinderFactory.create(null, null, Collections.emptyMap(), new BioCodec()));

Opennlp uses < START > and < END > to customize labeling entities, and names entities are marked with colons after START, such as < START:person >

Parameter description


On the engineering level, using maxent is an excellent way of creating programs which perform very difficult classification tasks very well.


number of training iterations, ignored if -params is used.


minimal number of times a feature must be seen

Use model

After training the model above, the model can be used for analysis.

      NameFinderME nameFinder = new NameFinderME(nameFinderModel);

        // now test if it can detect the sample sentences

        String[] sentence = "NATO United States Barack Obama".split("\\s+");

        Span[] names = nameFinder.find(sentence);

                .forEach(span -> {
                    String named = IntStream.range(span.getStart(),span.getEnd())
                            .mapToObj(i -> sentence[i])
                            .collect(Collectors.joining(" "));
                    System.out.println("find type: "+ span.getType()+",name: " + named);

The output is as follows:

find type: organization,name: NATO
find type: location,name: United States
find type: person,name: Barack Obama


The annotation of custom named entities in opennlp gives a certain customization space, which is convenient for developers to customize special named entities in their respective fields, so as to improve the accuracy of word segmentation for specific named entities.