EduBehaviors-Kit

EduBehaviors-kit is a Python package for annotating educational dialogues with observable behavioral indicators and using them to model more complex educational constructs. It implements the EduBehaviors Framework , combining LLM-generated annotations, efficient encoders, and interpretable machine learning to support scalable, low-cost analysis of student–educator conversations.

This toolkit complements EduBehaviors-Studio, an LLM-assisted interface for iterative, auditable development of assertion-based schemas from a construct codebook or definition and corpus annotation. Studio exports LLM-annotated assertions as CSV files, which can be used to train custom encoders or combined with the kit’s existing behavioral features for construct modeling.

Links to the framework paper and EduBehaviors-Studio repository will be added when available.

• GitHub: https://github.com/scale-nssa/edubehaviors-kit/
• PyPI package: https://pypi.org/project/edubehaviors-kit/
• Created by: Xander Beberman, Julian Bernado, Ana T. Ribeiro, and Susanna Loeb at The SCALE Initiative at Stanford University.
• Free software: MIT License

Features

• WordAnnotator — counts or flags a word list in text
• AssertionAnnotator — scores text against the EduBehaviors assertions with their SetFit models
• standard_classifier — ridge logistic regression with sensible defaults
• ClassificationPipeline — annotate, split, train and evaluate a labelled dataframe in one step

Assertions

See the assertions page for the list of published assertion classifiers with their cross-LLM agreement and test F1 scores.

Usage

Quickstart

Demo script

import pandas as pd

from edubehaviors import ClassificationPipeline

data = pd.read_csv("examples/talkmoves_tutor.csv")
pipeline = ClassificationPipeline(
    data,
    words="all",
    assertions=["sentence_has_a_question"],
    label_column="label_press_for_reasoning",
    group_column="transcript",
    random_state=2026_09_17,
)

print(pipeline.report())

See the usage docs for details.

Documentation

Documentation is built with Zensical and deployed to GitHub Pages.

• Live site: https://scale-nssa.github.io/edubehaviors-kit/
• Preview locally: just docs-serve (serves at http://localhost:8000)
• Build: just docs-build

API documentation is auto-generated from docstrings using mkdocstrings.

Docs deploy automatically on push to main via GitHub Actions. To enable this, go to your repo's Settings > Pages and set the source to GitHub Actions.

Development

To set up for local development:

# Clone your fork
git clone git@github.com:your_username/edubehaviors-kit.git
cd edubehaviors-kit

# Install the project and its dev dependencies
uv sync

This installs the package in editable mode, so any changes you make to the source code are picked up immediately.

Run tests:

uv run pytest

Run quality checks (format, lint, type check, test):

just qa

Author

EduBehaviors-Kit was created in 2026 by Xander Beberman, Julian Bernado, and Ana T. Ribeiro at The SCALE Initiative at Stanford University.

Built with Cookiecutter and the audreyfeldroy/cookiecutter-pypackage project template.