README

---
language:
- tr
task_categories:
- sentence-similarity
- text-classification
license: mit
size_categories:
- n<1K
---

# Semantic Textual Similarity (STS) Dataset (Turkish)

This repository contains a Turkish **Semantic Textual Similarity (STS)** dataset created as part of a university assignment on semantic similarity, sentence embeddings, and vector representations in Natural Language Processing (NLP).

## Authors

- Muhammet Enes Nas
- Salih Dede

## About

The purpose of this project was to gain practical experience with:

- Semantic Textual Similarity (STS)
- Sentence Embeddings
- Vector Representations
- Similarity Scoring
- NLP Dataset Preparation

The dataset consists of Turkish sentence pairs annotated with a semantic similarity score ranging from **0 to 100**, where:

| Score | Meaning |
|-------:|---------|
| 0 | Completely unrelated |
| 25 | Weak semantic relation |
| 50 | Moderate similarity |
| 75 | High similarity |
| 100 | Semantically equivalent |

## Dataset Format

Each sample contains three fields:

| Column | Description |
|---------|-------------|
| `sentence1` | First sentence |
| `sentence2` | Second sentence |
| `humanScore` | Human-annotated semantic similarity score (0–100) |

Example:

```json
{
  "sentence1": "Doktor ameliyata geç kaldı.",
  "sentence2": "Ameliyatı yapacak hekim gecikti.",
  "humanScore": 97.85
}
```

## Intended Use

This dataset is intended for educational and research purposes, including:

- Sentence Embedding models
- Semantic Similarity prediction
- Siamese Networks
- SBERT fine-tuning
- Embedding evaluation
- NLP coursework

## License

MIT License.