SentiPers Dataset

Dataset Description

SentiPers is a Persian sentiment analysis dataset containing text samples labeled with sentiment polarity scores ranging from -2 (very negative) to 2 (very positive).

Original Repository: https://github.com/phosseini/SentiPers

Fork: https://github.com/k-forghani/SentiPers

Dataset Structure

Data Splits

The dataset is split into three subsets with fixed random seed (42) for reproducibility:

Split	Size	Percentage
Train	10,820	80%
Validation	1,352	10%
Test	1,353	10%

Data Fields

• text: The input text in Persian
• label: Sentiment polarity label (integer from 0 to 4)

Label Mapping

The original polarity scores are mapped to integer labels as follows:

Original Polarity	Label ID	Sentiment
-2	0	Very Negative
-1	1	Negative
0	2	Neutral
1	3	Positive
2	4	Very Positive

Dataset Creation

The dataset is preprocessed from sentipers.xlsx with the following steps:

1. Extract text and polarity columns
2. Remove duplicate texts (keeping first occurrence)
3. Map polarity scores to integer labels
4. Perform stratified splitting to maintain label distribution across splits

Loading the Dataset

import pandas as pd

train_df = pd.read_csv('data/train.csv')
val_df = pd.read_csv('data/val.csv')
test_df = pd.read_csv('data/test.csv')

Preprocessing Script

Run preprocess_data.py to regenerate the splits from the source file:

python preprocess_data.py

Citation

If you use this dataset, please cite the original SentiPers repository:

Original: https://github.com/phosseini/SentiPers

Fork: https://github.com/k-forghani/SentiPers