Chunk Parser

CRFChunk orchidpp

v0.2

Model Details

• Developer: Wannaphong Phatthiyaphaibun
• This report author: Wannaphong Phatthiyaphaibun
• Model date: 2021-01-21
• Model version: 0.2
• Used in PyThaiNLP version: 2.3
• GitHub: https://github.com/PyThaiNLP/pythainlp/pull/524
• License: CC0
• train notebook: https://github.com/PyThaiNLP/pythainlp_notebook/pull/1
• Dataset: ORCHID++ from Thai Treebanks Dataset. We extract sentence subtree from tree to train data. (5,000 tree up to 5,935 tree)

Intended Use

• Parser thai sentence to phrase structure
• Not suitable for other languages or other domains of orchid corpus

Factors

• Based on thai chunk parser problems.

Metrics

• Evaluation metrics include precision, recall and f1-score.

Training Data

ORCHID++ (90%) from Thai Treebanks Dataset

Evaluation Data

ORCHID++ (10%) from Thai Treebanks Dataset

Quantitative Analyses

              precision    recall  f1-score   support

        B-NP       0.95      0.98      0.96       518
        I-NP       0.86      0.91      0.88      2128
           O       0.87      0.91      0.89       280
        B-PP       0.91      0.77      0.83        65
        I-PP       0.66      0.52      0.59       252
         B-S       0.65      0.49      0.56        90
         I-S       0.67      0.49      0.56      1082
        B-VP       0.86      0.89      0.88       515
        I-VP       0.90      0.94      0.92      4565

   micro avg       0.86      0.86      0.86      9495
   macro avg       0.81      0.77      0.79      9495
weighted avg       0.86      0.86      0.86      9495
 samples avg       0.86      0.86      0.86      9495

Ethical Considerations

• It trains from the orchid++ corpus. It is possible to have a bias from the orchid++ corpus.

Caveats and Recommendations

• 1 Thai sentence with [(word,part-of-speech)] (part-of-speech model trained from orchid corpus)