devpatel18042004/Indian-captions

🇮🇳 Indian Captions Dataset

A large-scale multilingual Instagram caption dataset designed for Natural Language Processing (NLP), sentiment analysis, text classification, multilingual language modeling, social-media analytics, caption generation, and AI/ML research.

The dataset contains 105,000+ unique Instagram-style captions covering multiple languages, writing styles, emotions, sentiments, and social-media categories commonly used in Indian and multilingual social-media contexts.

---

📊 Dataset Overview

Property	Details
Dataset Name	Indian Captions Dataset
Dataset ID	devpatel18042004/Indian-captions
Total Records	105,000+
Primary Domain	Instagram / Social Media
Data Type	Text
Languages	English, Hindi, Gujarati and mixed/romanized forms
Main Task	NLP / Text Classification / Caption Analysis
Sentiment Labels	Positive, Neutral, Negative
Format	Tabular
License	See License section
Creator	Dev Patel

---

🎯 Purpose

The goal of this dataset is to provide a diverse collection of Instagram-style captions that can be used to develop and evaluate NLP and Generative AI systems.

The dataset includes captions representing different:

• 💬 Writing styles
• ❤️ Emotions
• 😊 Sentiments
• 📝 Caption lengths
• 🌐 Languages
• 🔥 Social-media trends
• #️⃣ Hashtag patterns
• 🎭 Personality and expression styles

It is particularly useful for research and experimentation involving Indian-language and code-mixed social-media text.

---

🌐 Languages

The dataset contains multilingual and mixed-language captions, including:

• 🇬🇧 English
• 🇮🇳 Hindi (Devanagari)
• 🇮🇳 Hindi / Hinglish
• 🇮🇳 Gujarati
• 🇮🇳 Gujarati / Gujlish
• 🔀 Mixed-language social-media text

The dataset intentionally includes both native-script and Romanized forms where applicable.

---

🏷️ Dataset Categories

Captions are organized into different semantic and social-media categories.

Examples include:

• Motivation
• Attitude
• Romantic
• Sad
• Happy
• Shayari
• Friendship
• Love
• Inspirational
• Emotional
• Cool
• Funny
• Lifestyle
• Confidence
• Festival
• Travel
• Self-Love
• Success
• Nature
• Celebration
• Social Media
• and other related caption styles

The exact category distribution may vary across the dataset.

---

😊 Sentiment

Each caption contains a sentiment label.

Available sentiment classes include:

• Positive
• Neutral
• Negative

Example:

Caption:
"Work hard today and build the life you dream about. 🚀"

Sentiment:
Positive

---

📏 Caption Length

Captions are categorized using a length tier.

Possible values include:

• Short
• Medium
• Long

The dataset also provides numerical measurements:

• character_count
• word_count

These fields make the dataset useful for studying relationships between caption length, sentiment, category, and engagement-oriented writing styles.

---

#️⃣ Hashtags

Many captions contain associated hashtags.

Example:

#motivation #success #goals #mindset #growth

The hashtag field can be used for:

• Hashtag analysis
• Topic classification
• Social-media trend analysis
• Caption recommendation
• Hashtag generation
• Text-to-hashtag systems

---

📋 Dataset Schema

Each record contains the following fields:

Column	Type	Description
caption_id	string	Unique identifier for the caption
caption_text	string	Actual caption text
language	string	Language or language variant
category	string	Semantic/social-media category
length_tier	string	Short, Medium, or Long
sentiment	string	Positive, Neutral, or Negative
hashtags	string	Associated hashtags
character_count	integer	Number of characters in the caption
word_count	integer	Number of words in the caption
has_emoji	boolean	Whether the caption contains an emoji

---

🧾 Example Records

caption_id:
CAP_000001

caption_text:
"Rise: Work hard in peace today 🚀"

language:
English

category:
Motivation

length_tier:
Short

sentiment:
Positive

hashtags:
#ambition #growthmindset #goals #discipline #dreambig

character_count:
32

word_count:
7

has_emoji:
True

Another example:

caption_id:
CAP_000003

caption_text:
"Yaad aavi fari — vaat puri. 💧"

language:
Gujarati (Gujlish)

category:
Sad

length_tier:
Short

sentiment:
Negative

has_emoji:
True

---

🚀 Intended Use

This dataset can be used for a wide range of NLP and Generative AI applications.

1. Sentiment Analysis

Train models to classify social-media captions into:

Positive
Neutral
Negative

2. Text Classification

Classify captions according to their semantic or social-media category.

Example:

Input:
"Never stop believing in yourself."

Output:
Motivation

3. Multilingual NLP

The dataset can be used to experiment with multilingual and code-mixed Indian-language NLP.

Potential applications include:

• Hindi NLP
• Gujarati NLP
• Hinglish NLP
• Gujlish NLP
• Cross-lingual classification
• Language identification

4. Caption Generation

The dataset can be used as training or evaluation data for systems that generate Instagram-style captions.

Example:

Input:
Theme = Motivation
Language = Gujarati

Output:
A motivational Gujarati caption.

5. Caption Recommendation

Build systems that recommend captions based on:

• Mood
• Category
• Language
• Sentiment
• Caption length

6. Hashtag Recommendation

Use caption text and category information to develop hashtag recommendation systems.

7. Social Media Analytics

The dataset can support research into:

• Language usage
• Sentiment trends
• Caption styles
• Emoji usage
• Hashtag patterns
• Multilingual social-media communication

---

🤖 Possible Machine Learning Tasks

The dataset can be adapted for several supervised learning problems.

Sentiment Classification

Input:
caption_text

Target:
sentiment

Category Classification

Input:
caption_text

Target:
category

Language Classification

Input:
caption_text

Target:
language

Emoji Prediction

Input:
caption_text

Target:
has_emoji

Caption Length Prediction

Input:
caption_text

Target:
length_tier

Multi-Task Learning

Multiple targets can also be predicted simultaneously:

caption_text
      ↓
   NLP Model
      ↓
 ┌────┼───────────┐
 ↓    ↓           ↓
Language Category Sentiment
             ↓
          Emoji

---

🧠 Recommended Models

The dataset can be used with traditional machine-learning algorithms as well as modern Transformer models.

Traditional ML

• Logistic Regression
• Naive Bayes
• Support Vector Machines
• Random Forest
• XGBoost
• LightGBM

Transformer Models

Potential models include:

• BERT
• RoBERTa
• DistilBERT
• XLM-RoBERTa
• IndicBERT
• MuRIL
• mBERT

Generative Models

The dataset can also be useful for experimentation with:

• Llama
• Qwen
• Gemma
• Mistral
• Indic-language LLMs
• Other instruction-tuned language models

---

📥 Loading the Dataset

You can load the dataset using the Hugging Face datasets library.

from datasets import load_dataset

dataset = load_dataset(
    "devpatel18042004/Indian-captions"
)

print(dataset)

Access the training split:

train = dataset["train"]

print(train[0])

---

🐍 Using Pandas

from datasets import load_dataset

dataset = load_dataset(
    "devpatel18042004/Indian-captions"
)

df = dataset["train"].to_pandas()

print(df.head())
print(df.shape)

---

🔎 Example Filtering

Get Gujarati captions

gujarati = df[
    df["language"].str.contains(
        "Gujarati",
        case=False,
        na=False
    )
]

print(gujarati.head())

Get positive captions

positive = df[
    df["sentiment"] == "Positive"
]

print(positive.head())

Get motivational captions

motivation = df[
    df["category"] == "Motivation"
]

print(motivation.head())

---

📈 Dataset Statistics

The dataset provides additional metadata for every caption.

These include:

character_count
word_count
has_emoji
length_tier
sentiment
language
category

This makes the dataset suitable for exploratory data analysis and feature engineering.

Possible analyses include:

• Average caption length by language
• Sentiment distribution
• Category distribution
• Emoji usage by category
• Hashtag frequency
• Language distribution
• Short vs. medium vs. long captions
• Sentiment by language
• Category vs. sentiment

---

🧹 Data Quality

The dataset includes structured metadata intended to make downstream processing easier.

Before using the dataset for production or model training, users should consider performing additional preprocessing such as:

• Duplicate detection
• Text normalization
• Unicode normalization
• Hashtag normalization
• Emoji normalization
• Language verification
• Category validation
• Sentiment validation
• Train/validation/test splitting

For machine-learning experiments, avoid allowing near-duplicate captions to appear across different splits because this can cause data leakage and artificially inflated evaluation scores.

---

⚠️ Limitations

This dataset has several important limitations.

Synthetic / Curated Nature

The captions may include generated or curated social-media-style text and should not automatically be interpreted as naturally occurring Instagram posts.

Cultural Representation

Although the dataset focuses on Indian and multilingual caption styles, it should not be considered a statistically representative sample of all Indian social-media users.

Language Classification

Some captions are multilingual, code-mixed, or written using Romanized Indian languages. Therefore, language labels may not always correspond to a single linguistic variety.

Sentiment Subjectivity

Sentiment is inherently subjective. Some captions may reasonably be interpreted differently by different annotators or models.

Category Overlap

A caption may naturally belong to multiple categories. The provided category should therefore be treated as a dataset label rather than an absolute semantic truth.

---

🔐 Privacy

This dataset is intended to contain caption text and associated metadata rather than personally identifiable user information.

Users of this dataset should not attempt to infer or reconstruct the identity of individuals from caption content.

If this dataset is extended using real-world social-media data, appropriate privacy, consent, platform-policy, and legal requirements should be followed.

---

⚖️ Ethical Considerations

Potential applications should consider:

• Bias in language and sentiment labels
• Cultural stereotypes
• Misclassification of code-mixed language
• Representation imbalance
• Automated content generation
• Responsible use of social-media text

The dataset should not be used to make high-impact decisions about individuals.

---

📜 License

Please refer to the license specified in the Hugging Face repository.

If you redistribute, modify, or build upon this dataset, follow the applicable license requirements and provide appropriate attribution.

---

👨‍💻 Dataset Creator

Dev Patel

Hugging Face:

devpatel18042004

Dataset:

devpatel18042004/Indian-captions

---

🤝 Contributions

Contributions and improvements are welcome.

Potential contributions include:

• Additional Indian-language captions
• Improved language labels
• Better category annotations
• Sentiment validation
• Duplicate detection
• Dataset quality improvements
• Additional metadata
• Benchmark datasets
• Evaluation scripts

---

📌 Citation

If you use this dataset in a research project, application, article, or experiment, please cite the Hugging Face dataset.

@dataset{dev_patel_indian_captions,
  author       = {Dev Patel},
  title        = {Indian Captions Dataset},
  publisher    = {Hugging Face},
  year         = {2026},
  url          = {https://huggingface.co/datasets/devpatel18042004/Indian-captions}
}

---

⭐ Acknowledgement

If this dataset is useful for your project, consider starring the repository on Hugging Face and citing the dataset in your work.

---

🔗 Dataset

Hugging Face: https://huggingface.co/datasets/devpatel18042004/Indian-captions

---

🏷️ Keywords

Instagram Captions Indian Languages Hindi Gujarati Hinglish Gujlish NLP Natural Language Processing Sentiment Analysis Text Classification Multilingual NLP Social Media Caption Generation Hashtag Recommendation Generative AI Machine Learning Deep Learning Indian NLP