---
license: apache-2.0
---
# Semantic Integrity Analysis Dataset

## Overview

This dataset is designed for detecting semantic integrity violations between sentence pairs.

Each data instance contains two sentences and a label indicating the semantic relationship between them.

The dataset supports multi-class text pair classification.

---

## Task Description

Given two sentences (sentence1 and sentence2), the model must classify the relationship as:

- 0 → Contradiction
- 1 → Inconsistency
- 2 → Duplication

This task is similar to Natural Language Inference (NLI), but focuses on semantic validation within structured documents.

---

## Dataset Structure

Each row contains:

- sentence1 (string)
- sentence2 (string)
- label (integer)

Example:

sentence1: "The report was submitted in 2022."
sentence2: "The report was submitted in 2023."
label: 1

---

## Label Description

| Label | Category        | Meaning |
|-------|-----------------|---------|
| 0     | Contradiction   | Opposite meaning between sentences |
| 1     | Inconsistency   | Conflicting details or mismatched facts |
| 2     | Duplication     | Same or nearly same meaning |

---

## Data Source

The dataset was created from four structured documents (doc1, doc2, doc3, doc4).

Sentence pairs were extracted and manually annotated.

---

## Annotation Process

Annotation was performed manually based on semantic relationship guidelines.

Each sentence pair was reviewed and labeled into one of three categories.

---

## Intended Use

This dataset can be used for:

- Fine-tuning transformer models
- Semantic validation systems
- Document integrity checking
- NLP research on sentence-pair classification

---

## Limitations

- Limited dataset size
- Domain-specific content may reduce generalization
- Manual annotation may introduce bias

---

## Ethical Considerations

The dataset does not contain sensitive personal information.

It is intended for research and educational use only.