https://github.com/aydakikio/goemotion_persian
Persian translation of Google's GoEmotions dataset with 27 fine grained emotion
https://github.com/aydakikio/goemotion_persian
bert bert-fine-tuning emotion-detection emotion-recognition fa farsi fine-tuning goemotion-persian multi-label-classification nlp persian sentiment-analysis text-classification translated-dataset
Last synced: 6 days ago
JSON representation
Persian translation of Google's GoEmotions dataset with 27 fine grained emotion
- Host: GitHub
- URL: https://github.com/aydakikio/goemotion_persian
- Owner: aydakikio
- License: apache-2.0
- Created: 2026-05-26T07:26:27.000Z (about 1 month ago)
- Default Branch: main
- Last Pushed: 2026-06-05T17:55:41.000Z (21 days ago)
- Last Synced: 2026-06-05T19:23:45.685Z (21 days ago)
- Topics: bert, bert-fine-tuning, emotion-detection, emotion-recognition, fa, farsi, fine-tuning, goemotion-persian, multi-label-classification, nlp, persian, sentiment-analysis, text-classification, translated-dataset
- Language: Python
- Homepage:
- Size: 4.96 MB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
- Citation: CITATION.cff
Awesome Lists containing this project
README
# GoEmotion Persian
-blue.svg?style=for-the-badge)
## 📖 Overview
This dataset is a Persian (Farsi) translation of the **GoEmotions** benchmark,
originally created by Google Research. GoEmotions is one of the largest
fine-grained emotion datasets available, containing Reddit comments labeled
across 27 emotion categories plus Neutral.
This Persian version makes the GoEmotions benchmark accessible for **Persian
NLP research**, enabling tasks such as emotion detection, sentiment analysis,
and multi-label text classification in Farsi.
---
## 📥 Dataset Access
The GoEmotion Persian dataset can be accessed in multiple ways:
### 1️⃣ Kaggle Dataset
The dataset is also available on Kaggle for direct use in notebooks:
👉 [https://www.kaggle.com/datasets/aydakikio/goemotion-persian](https://www.kaggle.com/datasets/aydakikio/goemotion-persian)
You can use it directly in Kaggle:
```
import pandas as pd
path = "/kaggle/input/goemotion-persian/translated_train.tsv"
df = pd.read_csv(path, sep="\t", header=None)
```
---
### 2️⃣ Download from GitHub Releases
Pre-packaged dataset versions are available in the **Releases** section:
👉 [https://github.com/aydakikio/goemotion_persian/releases](https://github.com/aydakikio/goemotion_persian/releases)
You can download:
* Raw TSV files
* Translated dataset
---
### 3️⃣ Clone the Repository (Recommended for development)
```bash
git clone https://github.com/aydakikio/goemotion_persian.git
cd goemotion_persian
```
---
## 🧭 Provenance
This dataset is a Persian translation of the **GoEmotions benchmark** created by Google Research.
The dataset was constructed using a hybrid pipeline:
### 🤖 Automated Translation
* The majority of translations were generated using:
* **Gemini 2.5**
* **Gemini 2.5 Lite**
* These models were used to translate English Reddit comments into Persian while preserving emotional context and label alignment.
### ✋ Manual Correction
* A subset of samples was manually translated and corrected.
* Manual review focused on:
* untranslated or partially translated rows
* noisy inputs (URLs, special tokens, malformed text)
* improving fluency and cultural adaptation in Persian
### 🧪 Quality Control
* Automated validation tests were applied to ensure:
* Persian script consistency
* missing or empty translations detection
* duplication filtering
* length ratio sanity checks between source and target text
### 📊 Human Review
* A random sample of ~500 rows from each split was manually inspected to assess translation quality and consistency.
### 🔗 Reproducibility
This repository includes:
* translation pipeline implementation
* dataset preprocessing scripts
* QA / validation tools
* experiment and processing logs
* skipped / error batch tracking
---
## 🗂️ Repository Structure
```
goemotion_persian/
├── data
│ ├── raw
│ │ ├── raw_dev.tsv # 5,426 rows
│ │ ├── raw_test.tsv # 5,427 rows
│ │ └── raw_train.tsv # 43,410 rows
│ ├── test_sample
│ │ ├── dev_dataset_manual_review_200.csv #200 rows
│ │ ├── test_dataset_manual_review_200.csv #200 rows
│ │ └── train_dataset_manual_review_500.csv #500 rows
│ └── translated
│ ├── translated_dev.tsv # 5,426 rows
│ ├── translated_test.tsv # 5,427 rows
│ └── translated_train.tsv # 43,410 rows
├── source_code
│ ├── audiot_test.py #Quality tester
| └── translator.py #translation pipeline
├── log_files
│ ├── test
│ │ ├── dev_translate_test_1.log
│ │ ├── dev_translate_test_2.log
│ │ ├── dev_translate_test_3.log
│ │ ├── test_dataset_test_1.log
│ │ ├── test_dataset_test_2.log
│ │ ├── train_translate_test_1.log
│ │ ├── train_translate_test_2.log
│ │ └── train_translate_test_3.log
│ └── translation
│ ├── dev_translation_logs.log
│ ├── skipped_batches.log
│ ├── test_translation_logs.log
│ ├── train_translation_logs_day_1.log
│ └── train_translation_logs_day_2.log
├── README.md
└── CITATION.cff
```
---
## 🌞 Dataset Details
| Property | Value |
|---|---|
| Language | Persian (Farsi) |
| Original Language | English |
| Total Records | ~54,000 |
| Training Set | 43,410 |
| Validation Set | 5,426 |
| Test Set | 5,427 |
| Number of Labels | 27 emotions + Neutral |
| Format | TSV (tab-separated) |
| Translation Model | Gap GPT API |
---
## 🧭 Dataset Structure
Each file contains three columns with **no header row**:
| Column | Content |
|---|---|
| Column 0 | Persian translated text |
| Column 1 | Emotion label(s) |
| Column 2 | UUID |
---
## 💻 How to Use
```
import pandas as pd
# Load splits
train = pd.read_csv("data/train.tsv", sep="\t", header=None,
names=["text", "labels", "id"])
dev = pd.read_csv("data/dev.tsv", sep="\t", header=None,
names=["text", "labels", "id"])
test = pd.read_csv("data/test.tsv", sep="\t", header=None,
names=["text", "labels", "id"])
print(train.head())
```
> [!NOTE]
> Labels are stored as comma-separated emotion names (e.g. `joy,admiration`).
> A single text can have multiple labels.
---
## 🌿 Emotion Categories
`admiration` `amusement` `anger` `annoyance` `approval` `caring` `confusion`
`curiosity` `desire` `disappointment` `disapproval` `disgust` `embarrassment`
`excitement` `fear` `gratitude` `grief` `joy` `love` `nervousness` `optimism`
`pride` `realization` `relief` `remorse` `sadness` `surprise` `neutral`
---
## ⚓ GoEmotions Label Codes (Official)
| ID | Emotion | ID | Emotion |
|---|---|---|---|
| 0 | admiration | 14 | fear |
| 1 | amusement | 15 | gratitude |
| 2 | anger | 16 | grief |
| 3 | annoyance | 17 | joy |
| 4 | approval | 18 | love |
| 5 | caring | 19 | nervousness |
| 6 | confusion | 20 | optimism |
| 7 | curiosity | 21 | pride |
| 8 | desire | 22 | realization |
| 9 | disappointment | 23 | relief |
| 10 | disapproval | 24 | remorse |
| 11 | disgust | 25 | sadness |
| 12 | embarrassment | 26 | surprise |
| 13 | excitement | 27 | neutral |
> [!NOTE]
> These are multi label indices, a single text can have multiple active labels.
---
## 🔨 Quality Assurance
The translation pipeline included:
- Automated audit tests (Persian script validation, missing translation detection, length ratio checks, duplicate detection)
- Manual review of a 500-row random sample
- Exception handling for untranslatable rows (URLs, special tokens)
---
## ⚠️ Limitations
- Translations were generated via the **GAP GPT API** and are **not human-verified** beyond the 500-row sample review.
- Some idiomatic Reddit expressions may not translate accurately into Persian.
- Inherited label noise from the original GoEmotions dataset applies.
---
## 🔔 Original Dataset
This dataset is based on:
> Demszky, D., Movshovitz-Attias, D., Ko, J., Cowen, A., Nemade, G., & Ravi, S. (2020).
> **GoEmotions: A Dataset of Fine-Grained Emotions.**
> *Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL 2020).*
> [https://arxiv.org/abs/2005.00547](https://arxiv.org/abs/2005.00547)
Original dataset links:
- GitHub: [google-research/goemotions](https://github.com/google-research/google-research/tree/master/goemotions)
- Kaggle: [debarshichanda/goemotions](https://www.kaggle.com/datasets/debarshichanda/goemotions)
---
## 🫖 Citation
If you use this dataset in your research, please cite both the original GoEmotions paper and this dataset.
**This dataset:**
```
@dataset{ayda_khoshkan_2026,
title={GoEmotion Persian},
url={https://www.kaggle.com/dsv/16646751},
DOI={10.34740/KAGGLE/DSV/16646751},
publisher={Kaggle},
author={Ayda Khoshkan},
year={2026}
}
```
**Original GoEmotions paper:**
```
@inproceedings{demszky-etal-2020-goemotions,
title = {{G}o{E}motions: A Dataset of Fine-Grained Emotions},
author = {Demszky, Dorottya and Movshovitz-Attias, Dana and Ko, Jeongwook
and Cowen, Alan and Nemade, Gaurav and Ravi, Sujith},
booktitle = {Proceedings of the 58th Annual Meeting of the Association for
Computational Linguistics},
year = {2020},
url = {https://arxiv.org/abs/2005.00547}
}
```
---
## 🪶 License
This dataset is released under **[CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)** and **Apache 2.0**.
The original GoEmotions dataset is licensed under **[CC0](https://creativecommons.org/publicdomain/zero/1.0/)** and **Apache 2.0** by Google Research.