LoveJesus/intertextual-classifier-chirho

Intertextual Classifier (Chirho)

RoBERTa-base fine-tuned for classifying biblical cross-reference connection types.

"For God so loved the world that he gave his only begotten Son, that whoever believes in him should not perish but have eternal life." - John 3:16

Model Description

Given two Bible passages that are cross-referenced, this model classifies the type of intertextual connection between them into one of 7 categories:

Label	Description
thematic_parallel	Passages share the same theme or topic
direct_quote	One passage directly quotes another
prophetic_fulfillment	OT prophecy fulfilled in NT
typological	OT type foreshadowing NT antitype
contrast	Passages present contrasting ideas
historical_narrative	Shared historical events or figures
theological_expansion	Later passage expands on earlier theology

Training Details

• Base model: roberta-base (125M params)
• Training data: 19,164 balanced examples (Grok-labeled from TSK cross-references)
• Class balancing: WeightedTrainer with inverse-frequency CrossEntropyLoss + majority class capping
• Epochs: 8
• Best epoch: 8 (by eval loss)

Metrics (v2 - Retrained Feb 2026)

Metric	Value
Macro F1	0.761
Micro F1	0.853
Precision	0.665
Recall	0.939
Eval Loss	0.501

Improvement over v1

Metric	v1 (Original)	v2 (Retrained)	Change
Macro F1	0.42	0.761	+81%
Micro F1	0.72	0.853	+18%

Root cause of v1 weakness: 76% class imbalance (thematic_parallel dominated). Fixed with:

1. Balanced dataset (cap majority class, keep all minority examples)
2. WeightedTrainer with inverse-frequency class weights

Usage

from transformers import pipeline

classifier = pipeline(
    "text-classification",
    model="LoveJesus/intertextual-classifier-chirho",
    top_k=None,
)

text = "[CLS] Genesis 3:15 And I will put enmity between thee and the woman, and between thy seed and her seed; it shall bruise thy head, and thou shalt bruise his heel. [SEP] Galatians 4:4 But when the fulness of the time was come, God sent forth his Son, made of a woman, made under the law [SEP]"

result = classifier(text)
print(result)
# [{'label': 'prophetic_fulfillment', 'score': 0.95}, ...]

Part of Bible ML Pipeline

This model is part of the Intertextual Reference Network pipeline:

1. Embedder (LoveJesus/intertextual-embedder-chirho): Finds similar passages
2. Classifier (this model): Classifies the connection type

Dataset: LoveJesus/intertextual-dataset-chirho