README.md

metadata

dataset_info:
  features:
    - name: text
      dtype: string
    - name: label
      dtype: string
    - name: source
      dtype: string
    - name: domain
      dtype: string
    - name: language
      dtype: string
  splits:
    - name: train
      num_bytes: 1364685913
      num_examples: 3147478
    - name: validation
      num_bytes: 170841288
      num_examples: 393435
    - name: test
      num_bytes: 170338153
      num_examples: 393436
  download_size: 988308759
  dataset_size: 1705865354
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
license: apache-2.0
task_categories:
  - text-classification
language:
  - ar
  - de
  - en
  - es
  - fr
  - hi
  - id
  - it
  - ko
  - ms
  - pt
  - ru
  - tr
  - vi
  - zh
  - ja
tags:
  - sentiment
  - multilingual
  - emotion
  - review
  - classification
pretty_name: text
size_categories:
  - 1M<n<10M

Overview

MultilingualSentiment is a sentiment classification dataset that encompasses three sentiment labels: Positive, Neutral, Negative

The dataset spans multiple languages and covers a wide range of domains, making it ideal for multilingual sentiment analysis tasks.

Dataset Information

The dataset was meticulously collected and aggregated from various sources, including Hugging Face and Kaggle. These sources provide diverse languages and domains to ensure a comprehensive and balanced dataset.

• Total records: 3,934,349
• The dataset is divided into three subsets: train, validation, and test, with a ratio of 8:1:1:
  • Train: 3,147,478
  • Validation: 393,435
  • Test: 393,436

Number of Records per Language

Language	Count
Arabic (ar)	208,375
German (de)	212,853
English (en)	1,519,860
Spanish (es)	222,911
French (fr)	262,645
Hindi (hi)	9,423
Indonesian (id)	12,536
Italian (it)	3,020
Japanese (ja)	335,656
Korean (ko)	259,998
Malay (ms)	6,661
Multilingual	9,391
Portuguese (pt)	49,188
Russian (ru)	205,186
Turkish (tr)	44,743
Vietnamese (vi)	127,068
Chinese (zh)	444,835

Number of Records per Label

Label	Count
Negative	1,436,539
Neutral	1,041,512
Positive	1,456,298

Applications

This dataset is well-suited for training and evaluating models in multilingual sentiment analysis, natural language processing (NLP), and domain-specific sentiment classification tasks.

Loading dataset

from datasets import load_dataset

# Load the MultilingualSentiment dataset
dataset = load_dataset("clapAI/MultiLingualSentiment")

print(dataset)

DatasetDict({
    train: Dataset({
        features: ['text', 'label', 'source', 'domain', 'language'],
        num_rows: 3147478
    })
    validation: Dataset({
        features: ['text', 'label', 'source', 'domain', 'language'],
        num_rows: 393435
    })
    test: Dataset({
        features: ['text', 'label', 'source', 'domain', 'language'],
        num_rows: 393436
    })
})

Citation

@dataset{clapAI2024multilingualsentiment,
  title        = {MultilingualSentiment: A Multilingual Sentiment Classification Dataset},
  author       = {clapAI},
  year         = {2024},
  url          = {https://huggingface.co/datasets/clapAI/MultiLingualSentiment},
  description  = {A multilingual dataset for sentiment analysis with labels: positive, neutral, negative, covering diverse languages and domains.},
}