Registry / ai-ml / underthesea-core

underthesea-core

library3.3.2pypypi✓ verified 81d ago

Core Python library for Vietnamese Natural Language Processing (NLP), providing tokenization, word segmentation, POS tagging, named entity recognition, sentiment analysis, and text classification. Current version 3.3.2 targets Python ≥3.10. Release cadence: irregular, with major updates every few months.

ai-ml

pip install underthesea-core

github ↗homepage ↗

INSTALL

IMPORT

SIG · UNDERTHESEA-CORE

underthesea-core

ai-mlpythonv3.3.2

harness data pending

Install & Compatibility

Where this runs

No compatibility data collected yet for this library.

Code

Verified usage

Verified import paths — ran on the pinned version, not inferred.

underthesea_core

✓ import underthesea_core

✗ import underthesea

Basic usage of underthesea core functions.

import underthesea # Word segmentation text = "Chào bạn, tôi là người Việt Nam." tokens = underthesea.word_tokenize(text) print(tokens) # POS tagging pos_tags = underthesea.pos_tag(text) print(pos_tags) # Named entity recognition entities = underthesea.ner(text) print(entities) # Sentiment analysis (placeholder: uses a simple model) sentiment = underthesea.sentiment(text) print(sentiment) # Text classification category = underthesea.classify("Công nghệ thông tin") print(category)

Debug

Known issues

breakingBreaking change in v3.0.0: The function `word_tokenize` no longer accepts `format` parameter. Use `word_tokenize(text, format='text')` is removed; only list output is supported.

fix

Remove `format` parameter from calls to `word_tokenize`.

affects: >=3.0.0

breakingBreaking change in v3.0.0: `pos_tag` returns a list of tuples instead of a string. Check downstream code expecting string output.

fix

Update code to process list of tuples returned by `pos_tag`.

affects: >=3.0.0

gotchaModels are automatically downloaded on first use. This may take time and requires internet. Offline usage requires manual model caching.

fix

Pre-download models using `underthesea.download_all()` when internet is available.

affects: all

deprecatedFunction `sentiment` uses an older, less accurate model. It is deprecated in favor of Hugging Face Transformers models.

fix

Use `from transformers import pipeline` and load a Vietnamese sentiment model instead.

affects: >=3.3.0

breakingIn v2.x, `classify` returned a string. In v3.x, it returns a dictionary with keys 'label' and 'score'. This breaks code expecting a string.

fix

Update code to access `classify(text)['label']` instead.

affects: >=3.0.0

Upgrade

Version history

3.3.2latest on PyPI · released May 24, 2026

Audit

Dependencies

numpyrequiredNumerical operations for model inference

transformersoptionalTransformer-based models for advanced NLP tasks (optional for some features)

Agent activity

19 hits · last 30 days

node

OpenAI (training)

Resources

githubhttps://github.com/undertheseanlp/underthesea/ ↗

homepagehttps://github.com/undertheseanlp/underthesea/tree/main/extensions/underthesea_core ↗