Dataset Card for KazakhTextDuplicates

Dataset Details

Dataset Description

The KazakhTextDuplicates dataset is a collection of Kazakh-language texts containing duplicates with different levels of modification. The dataset includes exact duplicates, contextual duplicates, and partial duplicates, making it valuable for research in text similarity, duplicate detection, information retrieval, and plagiarism detection.

• Developed by: Arailym Tleubayeva
• Language(s) (NLP): Kazakh (kk)
• License: CC-BY-4.0

Uses

Direct Use

This dataset can be used for:

• Duplicate text detection: Identifying identical or near-identical texts in large collections.
• Plagiarism detection: Detecting reworded or slightly modified texts.
• Information Retrieval (IR): Improving search engines by training models to recognize near-duplicate documents.
• Semantic Text Similarity (STS): Training NLP models to recognize different levels of text similarity.
• Data Augmentation: Generating paraphrased datasets for training models on low-resource Kazakh text data.

Out-of-Scope Use

• Not suitable for general language modeling without additional preprocessing.
• Not designed for machine translation tasks.

Dataset Structure

The dataset consists of Kazakh-language texts with three types of duplicate labels:

• Exact duplicate: The texts are identical.
• Contextual duplicate: The texts have been slightly modified but retain the same meaning.
• Partial duplicate: Only a portion of the text is similar.

Columns:

Column Name	Description
id	Unique document ID
content	Original Kazakh text
category	Text category (e.g., article, news, academic, etc.)
language	Language identifier (kk for Kazakh)
type_duplicate	Type of duplicate (exact, contextual, partial)
modified_content	Modified version of the original text

Dataset Creation

Curation Rationale

This dataset was created to address the lack of publicly available Kazakh-language datasets for duplicate detection and text similarity tasks. It is particularly useful for NLP applications involving Kazakh text retrieval, plagiarism detection, and duplicate content filtering.

Source Data

The dataset consists of curated text corpora from:

• Public Kazakh-language sources (news articles, academic texts, Wikipedia, etc.).
• Synthetic modifications created using NLP-based transformations.

Data Collection and Processing

The dataset was processed using:

• Text normalization (lowercasing, punctuation removal).
• Stopword removal (Kazakh stopwords).
• Duplication type classification (exact/contextual/partial).
• Data augmentation for modified versions.

Who are the source data producers?

• Kazakh-language content producers (media, academia).
• Synthetic text generation tools.

Annotations [optional]

Annotation process

The dataset was annotated using automatic similarity detection models and manual verification.

Who are the annotators?

The dataset was reviewed by Kazakh-speaking NLP researchers and linguists.

Personal and Sensitive Information

The dataset does not contain personal or sensitive data. Any potential identifiable information was removed during preprocessing.

Bias, Risks, and Limitations

• Bias in duplicate detection: The dataset might favor certain duplicate structures over others.
• Linguistic variation: Limited to standard Kazakh; regional dialects may not be fully represented.
• Machine-generated paraphrases: Some contextual duplicates were generated using NLP models, which may introduce biases.

Recommendations

Users should be aware of potential biases and perform additional validation when applying the dataset to real-world tasks.