README.md

---
annotations_creators:
- machine-generated
language_creators:
- found
language:
- en
license:
- mit
multilinguality:
- monolingual
size_categories:
- 100K<n<1M
source_datasets:
- original
task_categories:
- text-classification
- sentence-similarity
task_ids:
- semantic-similarity-classification
pretty_name: Wiki-727K
tags:
- text segmentation
- document segmentation
- topic segmentation
- topic shift detection
- semantic chunking
- chunking
- nlp
- wikipedia
dataset_info:
  features:
  - name: id
    dtype: string
  - name: ids
    sequence: string
  - name: sentences
    sequence: string
  - name: titles_mask
    sequence: uint8
  - name: levels
    sequence: uint8
  - name: labels
    sequence:
      class_label:
        names:
          '0': semantic-continuity
          '1': semantic-shift
  splits:
  - name: train
    num_bytes: 4754764877
    num_examples: 582160
  - name: validation
    num_bytes: 595209014
    num_examples: 72354
  - name: test
    num_bytes: 608033007
    num_examples: 73232
  download_size: 1569504207
  dataset_size: 5958006898
---
# Dataset Card for Wiki-727K Dataset

Wiki-727K is a large dataset for text segmentation, automatically extracted and labeled from Wikipedia. It is designed as a sentence-level sequence labeling task for identifying semantic or topic shift in documents.

## Dataset Overview

- **Train**: 582k
- **Validation**: 72k
- **Test**: 73k

## Features

- **id (string):** Document ID.
- **ids (sequence of string):** Sentence IDs for each document.
- **sentences (sequence of string):** Sentences in each document.
- **titles_mask (sequence of uint8):** Mask indicating if a sentence is a title (optional).
- **levels (sequence of uint8):** Hierarchical level of each sentence (optional).
- **labels (sequence of class):** Binary labels: `semantic-continuity` or `semantic-shift`.

## Usage

The dataset can be loaded using the HuggingFace `datasets` library:

```python
from datasets import load_dataset

titled_dataset = load_dataset('saeedabc/wiki727k', num_proc=8, trust_remote_code=True)

untitled_dataset = load_dataset('saeedabc/wiki727k', drop_titles=True, num_proc=8, trust_remote_code=True)

```

## Dataset Details

- **Homepage**: [Wiki-727K GitHub](https://github.com/koomri/text-segmentation)