CoolFace
Datasetpublic

shae2977/hinglish-youtube-sentiments-dataset

Hinglish YouTube Comments Sentiment Dataset A manually annotated dataset of 3,190 Hinglish YouTube comments for 3-class sentiment classification. Hinglish is the code-mixed Hindi-English language used by hundreds of millions of Indians online — written in Roman script, mixing Hindi and English words fluidly within the same sentence. This dataset was created because no sufficiently large, cleanly annotated Hinglish sentiment dataset existed for YouTube comment data specifically.… See the full description on the dataset page: https://huggingface.co/datasets/shae2977/hinglish-youtube-sentiments-dataset.

sourceHugging Facecc-by-4.0updated 4mo agoView on Hugging Face
0likes64downloads
Dataset Card

Hinglish YouTube Comments Sentiment Dataset

A manually annotated dataset of 3,190 Hinglish YouTube comments for 3-class sentiment classification. Hinglish is the code-mixed Hindi-English language used by hundreds of millions of Indians online — written in Roman script, mixing Hindi and English words fluidly within the same sentence.

This dataset was created because no sufficiently large, cleanly annotated Hinglish sentiment dataset existed for YouTube comment data specifically.

Dataset details

PropertyValue
Total samples3,190
ClassesNegative, Neutral, Positive
LanguageHinglish (code-mixed Hindi-English, Roman script)
SourceYouTube comments
DomainIndian general entertainment (Bollywood, music, comedy, cooking)
Annotation toolLabel Studio

Class distribution

ClassCountPercentage
Negative1,42744.73%
Neutral77024.14%
Positive99331.13%

Data collection

  • —Comments were scraped from YouTube videos in the Indian general entertainment space, including Bollywood, music, comedy, and cooking content.
  • —Comments were filtered to include only those written primarily in Roman script Hinglish. Pure English and pure Devanagari Hindi comments were excluded.

Annotation

Annotation was done manually using Label Studio. Each comment was labeled either Positive, Negative or Neutral.

Usage

python
from datasets import load_dataset

dataset = load_dataset("shae2977/hinglish-youtube-comments-dataset")

Related model

A fine-tuned XLM-RoBERTa model trained on this dataset is available at: shae2977/xlm-roberta-hinglish-sentiment-analysis

ModelWeighted F1
VADER (baseline)0.39
XLM-RoBERTa (fine-tuned on this dataset)0.67

Limitations

  • —Annotated by a single annotator — no inter-annotator agreement score available
  • —Domain is limited to general entertainment; may not generalize to sports, news, or politics
  • —Neutral class is underrepresented and harder to classify consistently
  • —No specific time period recorded for scraped comments

Author

Created by shae2977