README.md

---
# 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
language: en
license: mit
tags:
  - bible
  - chirho
  - intertextual
  - cross-reference
  - classification
  - roberta
  - bible-ml
datasets:
  - LoveJesus/intertextual-dataset-chirho
base_model: roberta-base
metrics:
  - f1
pipeline_tag: text-classification
---

# 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

```python
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](https://huggingface.co/spaces/LoveJesus/intertextual-reference-network-chirho) pipeline:

1. **Embedder** ([LoveJesus/intertextual-embedder-chirho](https://huggingface.co/LoveJesus/intertextual-embedder-chirho)): Finds similar passages
2. **Classifier** (this model): Classifies the connection type

Dataset: [LoveJesus/intertextual-dataset-chirho](https://huggingface.co/datasets/LoveJesus/intertextual-dataset-chirho)