Indonesia Instagram cyberbullying

"All" Dataset (Full Dataset)
This is the primary and most comprehensive dataset containing all text samples (comments). Each text sample has a multi-label label covering all possible categories, including neutral text (not cyberbullying) and various types of cyberbullying (e.g., neutral, flaming, denigration, racism, etc.).

Usage: This dataset is used in two scenarios:

Scenario A (Single-Stage Multi-Label Classification): The "All" dataset is used directly to train a model to classify text into one or more categories simultaneously (e.g., a text can be labeled neutral only, or both flaming and racism simultaneously).

Scenario B - Stage 1 (Binary Detection): This dataset is used to train a binary classification model. For this stage, the original multi-label labels are transformed into binary labels (Yes/No):

The binary label is No (Not Cyberbullying): If the text label is neutral.

Binary label value is Yes (Cyberbullying): If the text contains at least one cyberbullying type label (e.g., flaming, denigration, etc.).
Dataset Cyberbullying (Derived Dataset)
Definition: This is a subset of the "All Dataset." This dataset was created by filtering and sampling only text that was identified as cyberbullying (binary label value: Yes) in Scenario B - Phase 1.

Characteristics: This dataset no longer contains text with a neutral label. It only contains texts guaranteed to contain cyberbullying, along with a multi-label label detailing the type of cyberbullying (e.g., flaming, denigration, etc.).

Usage:

Scenario B - Phase 2 (Cyberbullying Type Classification): This dataset is used exclusively to train the model in the second phase of Scenario B. The goal is to classify the type of cyberbullying from a text, once the text has been confirmed as cyberbullying by the Phase 1 model.

全量数据集（All Dataset，完整数据集）
本数据集为核心且最全面的数据集，涵盖全部文本样本（即评论内容）。每条文本样本均带有覆盖全量类别的多标签标注，涵盖中性文本（非网络欺凌（cyberbullying））以及各类网络欺凌内容（例如中性、flaming、denigration、racism等）。

用途：本数据集可应用于两类场景：
场景A（单阶段多标签分类（Single-Stage Multi-Label Classification））：直接使用全量数据集训练模型，实现文本同时归入一个或多个类别（例如，某文本可仅标注为中性，或同时标注为flaming与racism）。

场景B-阶段1（二元检测（Binary Detection））：使用本数据集训练二元分类（binary classification）模型。此阶段需将原始多标签标注转换为二元标注（是/否）：
- 二元标注为“否（非网络欺凌）”：当文本标注为中性时。
- 二元标注为“是（网络欺凌）”：当文本包含至少一种网络欺凌类型标注（例如flaming、denigration等）。

网络欺凌数据集（Cyberbullying Dataset，衍生数据集）
定义：本数据集为全量数据集的子集，通过筛选并采样场景B-阶段1中被判定为网络欺凌（二元标注值为“是”）的文本构建而成。

特性：本数据集不再包含中性标注的文本，仅包含确定为网络欺凌的文本，并附带用于详细说明网络欺凌类型的多标签标注（例如flaming、denigration等）。

用途：
场景B-阶段2（网络欺凌类型分类（Cyberbullying Type Classification））：本数据集仅用于训练场景B的第二阶段模型，目标是在阶段1模型已确认文本属于网络欺凌的前提下，对文本的网络欺凌类型进行分类。