README.md

metadata

language: en
library_name: setfit
pipeline_tag: text-classification
base_model: sentence-transformers/paraphrase-mpnet-base-v2
datasets:
  - StanfordSCALE/assertions_llm_annotated_talkmoves
metrics:
  - f1
  - precision
  - recall
  - roc_auc
tags:
  - setfit
  - text-classification
  - education
  - classroom-discourse
  - edubehaviors
  - assertion
model-index:
  - name: assertion_references_student_work
    results:
      - task:
          type: text-classification
        dataset:
          name: assertions_llm_annotated_talkmoves
          type: StanfordSCALE/assertions_llm_annotated_talkmoves
          split: test
        metrics:
          - type: f1
            value: 0.5891
            name: F1 (positive class, test)
          - type: precision
            value: 0.6899
            name: Precision (positive class, test)
          - type: recall
            value: 0.5141
            name: Recall (positive class, test)
          - type: roc_auc
            value: 0.8862
            name: ROC-AUC (test)

Assertion: references student work

This classifier was trained for EduBehaviors: Assertion-based schemas for auditable dialogue coding and is usable through the Python package EduBehaviors-kit. This classifier was trained on an LLM-annotated subset of teacher utterances from the TalkMoves Dataset. See the Datasets section below for more information.

---

Training Details

Datasets

Dataset	Split	Size
StanfordSCALE/assertions_llm_annotated_talkmoves	train	13,183 (53.3%)
StanfordSCALE/assertions_llm_annotated_talkmoves	dev	3,294 (13.3%)
StanfordSCALE/assertions_llm_annotated_talkmoves	test	8,238 (33.3%)

This model's columns are assertion_references_student_work and split_references_student_work.

Base rate (share of rows labeled as True): 8.1% overall — 8.1% train, 8.1% dev, 8.2% test.

Labels and annotation

Labels were generated with LLM annotators. Krippendorff's alpha for this assertion is 0.665.

Hyperparameters

Parameter	Value
Base model (body)	sentence-transformers/paraphrase-mpnet-base-v2
Head	LogisticRegression
Body learning rate	2e-05
Head learning rate	0.01
Batch size	16 (contrastive phase) / 32 (head)
Epochs	10
Max steps	5000 (contrastive phase)
Eval max steps	100
Seed	20260904
Mixed precision	enabled on GPU

---

Evaluation

Results

Split	n	Base rate	Precision	Recall	F1 (positive class)	ROC-AUC	Average precision
dev	3,294	8.1%	0.681	0.594	0.635	0.904	0.647
test	8,238	8.2%	0.690	0.514	0.589	0.886	0.628

Limitations

• Labels come from LLM annotators, not human coders. Agreement between annotators with Krippendorff's Alpha is 0.665.
• Trained on teacher utterances only. Behaviour on student speech is untested.

---

How to Use

Message Structure

The model was trained on text built as:

{utterance}

The utterance is passed through as-is.

Running instructions

pip install setfit

from setfit import SetFitModel

model = SetFitModel.from_pretrained("StanfordSCALE/assertion_references_student_work")

text = 'Some of you built different models and I asked you a question about  that'
model.predict([text])        # -> array([1]) when the assertion holds
model.predict_proba([text])  # -> [[P(no), P(yes)]]

Citation

@misc{assertion_references_student_work,
  author = {Stanford SCALE Initiative},
  title  = {Assertion classifier: references student work},
  year   = {2026},
  url    = {https://huggingface.co/StanfordSCALE/assertion_references_student_work}
}