OPENNLP-1518: Roberta-based Models - Add support for utilization via Onxx

[rzo1]

For all changes:

• Is there a JIRA ticket associated with this PR? Is it referenced
  in the commit message?

• Does your PR title start with OPENNLP-XXXX where XXXX is the JIRA number you are trying to resolve? Pay particular attention to the hyphen "-" character.

• Has your PR been rebased against the latest commit within the target branch (typically main)?

• Is your initial contribution a single, squashed commit?

For code changes:

• Have you ensured that the full suite of tests is executed via mvn clean install at the root opennlp folder?
• Have you written or updated unit tests to verify your changes?
• If adding new dependencies to the code, are these dependencies licensed in a way that is compatible for inclusion under ASF 2.0?
• If applicable, have you updated the LICENSE file, including the main LICENSE file in opennlp folder?
• If applicable, have you updated the NOTICE file, including the main NOTICE file found in opennlp folder?

For documentation related changes:

• Have you ensured that format looks appropriate for the output in which it is rendered?

Note:

Converted the model mentioned in the issue to ONXX first. Did run a test via

public class RobertaGoEmotionsTest {

  private static final String MODEL_DIR = "roberta-base-go_emotions";

  public static void main(String[] args) throws Exception {

    final File modelDir = new File(MODEL_DIR);
    if (!modelDir.exists()) {
      System.err.println("Model directory not found: " + modelDir.getAbsolutePath());
      System.err.println("Clone it with: git clone https://huggingface.co/SamLowe/roberta-base-go_emotions");
      return;
    }

    final File onnxModel = new File(modelDir, "model.onnx");
    if (!onnxModel.exists()) {
      System.err.println("ONNX model file not found: " + onnxModel.getAbsolutePath());
      System.err.println("Download it from: https://huggingface.co/SamLowe/roberta-base-go_emotions/resolve/main/onnx/model.onnx");
      return;
    }

    final File vocabFile = new File(modelDir, "vocab.json");
    final File configFile = new File(modelDir, "config.json");

    // RoBERTa does not use token_type_ids
    final InferenceOptions inferenceOptions = new InferenceOptions();
    inferenceOptions.setIncludeTokenTypeIds(false);

    try (DocumentCategorizerDL categorizer = new DocumentCategorizerDL(
        onnxModel, vocabFile, configFile,
        new AverageClassificationScoringStrategy(),
        inferenceOptions)) {

      final String[] testSentences = {
          "I am so happy today!",
          "This is terrible and I hate it.",
          "Thank you so much for your help!",
          "I'm not sure what to think about this.",
          "That's hilarious, I can't stop laughing!"
      };

      for (final String sentence : testSentences) {
        System.out.println("Input: \"" + sentence + "\"");

        final SortedMap<Double, Set<String>> scores = categorizer.sortedScoreMap(new String[] {sentence});

        System.out.println("  Top emotions:");
        final List<Double> sortedScores = new ArrayList<>(scores.keySet());
        Collections.reverse(sortedScores);
        int count = 0;
        for (final Double score : sortedScores) {
          if (count++ >= 3) break;
          System.out.printf("    %-20s %.4f%n", scores.get(score), score);
        }
        System.out.println();
      }
    }
  }
}

This will fail with the unpatched version but will work with the patch:

Input: "I am so happy today!"
  Top emotions:
    [joy]                0,8805
    [excitement]         0,0736
    [neutral]            0,0105

Input: "This is terrible and I hate it."
  Top emotions:
    [anger]              0,6766
    [annoyance]          0,1317
    [disapproval]        0,1008

Input: "Thank you so much for your help!"
  Top emotions:
    [gratitude]          0,9992
    [approval]           0,0001
    [admiration]         0,0001

Input: "I'm not sure what to think about this."
  Top emotions:
    [confusion]          0,9703
    [neutral]            0,0082
    [disapproval]        0,0025

Input: "That's hilarious, I can't stop laughing!"
  Top emotions:
    [amusement]          0,7920
    [joy]                0,1405
    [neutral]            0,0279

Note: Since we do not use a JSON library (like jackson) here, I used a regex approach with a quirky check to parse a json based vocab file. The original issue failed due to missing Roberta special tokens.

[rzo1]

Let's wait with the merge until the current vote is completed.