Dataset assetOpen Source CommunityNatural Language ProcessingSarcasm Detection

CreativeLang/SARC_Sarcasm

This dataset is a large‑scale corpus for sarcasm research and for training and evaluating sarcasm detection systems. It contains 1.3 million sarcastic statements—ten times larger than any previous dataset—and a larger number of non‑sarcastic statements, enabling learning under both balanced and imbalanced label regimes. Each statement is self‑annotated (the sarcasm label is provided by the author rather than an external annotator) and includes user, topic, and dialogue context. The dataset’s accuracy has been evaluated, a sarcasm detection benchmark established, and baseline methods assessed.

Source

hugging_face

Created

Nov 28, 2025

Updated

Jul 11, 2023

Signals

247 views

Availability

Linked source ready

Overview

Dataset description and usage context

Dataset Overview

Dataset Name

Name: SARC_Sarcasm

Dataset Features

Feature List:
- text: string
- author: string
- score: int64
- ups: int64
- downs: int64
- date: string
- created_utc: int64
- subreddit: string
- id: string

Dataset Splits

Training Set:
- Samples: 12,704,751
- Size: 1,764,500,045 bytes

Dataset Size

Download Size: 903,559,115 bytes
Total Size: 1,764,500,045 bytes

License

License Type: cc-by-2.0

Dataset Description

Purpose: For sarcasm research and training/evaluating sarcasm detection systems
Scale: 1.3 million sarcastic statements, ten times larger than any previous dataset
Annotation Method: Self‑annotation by authors
Content: Includes user, topic, and dialogue context information
Evaluation & Benchmark: Accuracy evaluated, sarcasm detection benchmark established

Dataset Metadata

Type: Sarcasm
Task Type: Detection
Creation Year: 2018

Need downstream help?

Pair the dataset with AI analysis and content workflows.

Once the source passes your review, move straight into summarization, transformation, report drafting, or presentation generation with the JuheAI toolchain.

Explore AI studio