paralabelgen 0.1.1

A Python library for generating discrete paragraph labels from concept extraction, graph communities, and interpretable assignment rules.

paralabelgen

paralabelgen is a Python library for generating discrete multi-label annotations for text paragraphs from concept extraction, graph communities, and interpretable assignment rules.

• PyPI distribution: paralabelgen
• Python import package: labelgen
• Repository: https://github.com/HuRuilizhen/labelgen

Install

pip install paralabelgen
python -m spacy download en_core_web_sm

en_core_web_sm is the recommended default model. If you already use another compatible English spaCy pipeline, you can point spacy_model_name at that installed model instead.

Quick Start

from labelgen import LabelGenerator, LabelGeneratorConfig

paragraphs = [
    "OpenAI builds language models for developers.",
    "Developers use language models in production systems.",
]

config = LabelGeneratorConfig(
    use_nlp_extractor=False,
    use_graph_community_detection=False,
)
generator = LabelGenerator(LabelGeneratorConfig())
result = generator.fit_transform(paragraphs)

print("Concepts:")
for concept in result.concepts:
    print(concept.normalized, concept.kind, concept.document_frequency, sep=" | ")

print("Labels:")
for assignment in result.paragraph_labels:
    print(assignment.paragraph_id, assignment.label_ids, assignment.label_scores)

The default public pipeline uses spaCy extraction and Leiden community detection. Install the recommended spaCy model before running the example.

Public API

The main public entrypoints are:

• LabelGenerator
• LabelGeneratorConfig
• Paragraph, Concept, ConceptMention, Community, ParagraphLabels
• dump_result() and load_result()

Detailed API notes are available in docs/public_api.md.

Examples

Runnable examples are available in examples/:

• examples/basic_usage.py
• examples/custom_config.py
• examples/save_and_load.py

Configuration Notes

• fit() learns concept communities from a corpus.
• transform() applies previously learned communities to new paragraphs.
• fit_transform() learns and labels the same input in one pass.
• The default pipeline uses spaCy extraction and Leiden community detection.
• The default NLP path requires the configured spaCy model to be installed.
• en_core_web_sm is the recommended default model name.
• If the configured model is missing, the library raises an explicit runtime error.
• Set use_nlp_extractor=False to switch to the deterministic heuristic extractor.
• Set use_graph_community_detection=False to switch to deterministic connected-components community detection.
• The heuristic extractor uses capitalized spans as lightweight entities and non-stopword spans as candidate noun phrases.

Opt Out Of Enhanced Implementations

from labelgen import LabelGenerator, LabelGeneratorConfig

config = LabelGeneratorConfig(
    use_nlp_extractor=False,
    use_graph_community_detection=False,
)
generator = LabelGenerator(config)

Use A Different spaCy Model

from labelgen import LabelGenerator, LabelGeneratorConfig

config = LabelGeneratorConfig()
config.extraction.spacy_model_name = "en_core_web_md"

generator = LabelGenerator(config)