ChildGuard Dataset

A large-scale English dataset for child-targeted hate speech detection and analysis, designed to support child-focused mental health applications.

---

Overview

ChildGuard contains 351,877 annotated instances collected from Reddit, X (formerly Twitter), and YouTube. The dataset focuses exclusively on hate speech directed at children and incorporates age-specific annotations spanning:

• Younger Children (< 11)
• Pre-Teens (11–12)
• Teens (13–17)

To facilitate different analytical perspectives, ChildGuard is organized into lexical and contextual subsets.

---

Dataset Structure

Lexical Subset (lexical_childhate.csv)

Attribute	Value
Rows	157,280
Columns	text, actual_class, predicted_class, Hate

Description:
Focuses on vocabulary-level characteristics, including lexical patterns and sentiment cues associated with child-targeted hate speech. Age-group annotations are not included.

---

Contextual Subset (contextual_childhate.csv)

Attribute	Value
Rows	194,597
Columns	text, actual_class, predicted_class, Age_Group, Hate

Description:
Contains context-enriched instances with age-group annotations, supporting discourse-level and semantic analyses of child-targeted hate speech.

---

Full ChildGuard Dataset (childguard_dataset.csv)

Attribute	Value
Rows	351,877
Columns	text, actual_class, predicted_class, Age_Group, Hate

Description:
Integrates both lexical and contextual annotations into a single corpus for comprehensive analysis of child-targeted hate speech.

---

Dataset Statistics

Metric	Lexical	Contextual	ChildGuard
Instances	157,280	194,597	351,877
Unique Words	28,764	22,890	35,412
Average Text Length	19.1	22.7	21.2
Explicit Hate	26%	24%	25%
Implicit Hate	12%	14%	13%

---

Data Fields

Field	Description
text	Raw user-generated statement.
actual_class	Ground-truth label (Hate or Non-Hate).
predicted_class	Model-predicted label used for comparative evaluation.
Hate	Binary label (1 = Hate, 0 = Non-Hate).
Age_Group	Annotated age category (Younger Children, Pre-Teens, Teens). Available in contextual and full datasets only.

---

Dataset Files

The repository contains the following files:

File	Description
lexical_childhate.csv	Lexical subset containing vocabulary-level hate speech annotations.
contextual_childhate.csv	Contextual subset containing age-group annotations and discourse-level characteristics.
childguard_dataset.csv	Complete dataset integrating lexical and contextual annotations.