Persian Sentiment Analyzer for Product Reviews

📌 Overview

A sentiment analysis tool for Persian product reviews, classifying feedback into:

• ✅ Recommended
• ❌ Not Recommended
• 🤷 No Idea

✨ Features

• Preprocessing Pipeline
  Normalization, tokenization, stemming, and stopword removal for Persian text.
• Word2Vec Embeddings
  Custom-trained embeddings on 150K+ Persian reviews.
• Logistic Regression Classifier
  Achieves 75% accuracy on large datasets.
• Easy Prediction API
  predict_recommendation() function for new reviews.

📂 Dataset

I obtained a dataset of 150,000 samples from quera.org. Additionally, I acquired a larger dataset from Kaggle and included it in the repository as big_train.csv.

Loading the Dataset

To train the model using the smaller dataset:

train_data = pd.read_csv('train.csv')

For systems with powerful hardware to process the larger dataset:

train_data = pd.read_csv('big_train.csv', usecols=['body', 'recommendation_status'])

This combined dataset merges samples from both Quera and Kaggle sources.

Due to the large size of the big_train.csv file, I've split it into 12 RAR parts. To obtain the complete file, you only need to extract one of the parts.

A set of comments is provided in the comments.csv file which can be used as input for the model.

Downloading the big_train dataset from Hugging Face

To download the big_train dataset from Hugging Face, you can use the curl or wget tools. This dataset contains Persian commercial comments and is available as a CSV file.

Method 1: Using curl

curl -L -o big_train.csv "https://huggingface.co/datasets/2077devwave/persian_commercial_comments_filing/resolve/main/train.csv"

This command will save the file as train.csv in the current directory.

Method 2: Using wget

wget -O big_train.csv "https://huggingface.co/datasets/2077devwave/persian_commercial_comments_filing/resolve/main/train.csv"

If wget is not installed on your system, you can install it with the following command:

sudo apt install wget  # For Ubuntu and Debian
brew install wget  # For macOS

Method 3: Download with Python

You can use the requests library in Python:

import requests

url = "https://huggingface.co/datasets/2077devwave/persian_commercial_comments_filing/resolve/main/train.csv"
response = requests.get(url)

with open("big_train.csv", "wb") as file:
    file.write(response.content)

print("Download completed!")

With these methods, you can easily download and use the big_train dataset.

Hardware Specifications

This project was executed on hardware with the following specifications:

• 5 vCPU cores
• 12 GB RAM
• 15 GB GPU

If you face any issues with library installation or hardware requirements, you can use Google Collab as an alternative platform.

🛠️ Installation

1- Clone the repo:

git clone https://github.com/RezaGooner/Sentiment-Survey-Analyzer.git

2- Install dependencies:

pip install -r requirements.txt

🚀 Usage

from model import predict_recommendation

result = predict_recommendation("این گوشی ارزش خرید دارد")
print(result)

---

📊 Performance

Dataset Size Accuracy 150K (Quera) 60% +1000K (Combined) 75%

---

Real-World Test Cases

Below are sample reviews with star ratings, user recommendations, and model predictions for performance evaluation:

Review	Stars	User Recommended	Model Prediction	TP	TN	FP	FN
با توجه به قیمت بهترین انتخابه	⭐⭐⭐	✅	Recommended	1	0	0	0
کیفیت خوبی داره و برای پوست صورت گزینه مناسبی هست و از خریدم راضی هستم	⭐⭐⭐⭐⭐	✅	Recommended	1	0	0	0
عالیه.حیف که گرون شد	⭐⭐⭐	➖	Recommended	0	0	1	0
دستمال خوب و به صرفه ای میباشد .اما کیفیت آن نسبت به جعبه ای یک مقدار خشک است ولی تو شرایط بازار الان به صرفه میباشد. یک مقدار خشک میباشد	⭐⭐⭐	➖	No Opinion	0	0	0	0
با جلد پاره ارسال شد! اونم وسیله بهداشتی	⭐	❌	Not Recommended	0	1	0	0
قیمت رو دسمال ۱۴۰۰۰ تومن تولیدبرای هفته پیشه بعد ۱۷۰۰۰ میفروشن متاسفم	⭐	❌	Not Recommended	0	1	0	0
بنده چند بار خرید کردم ولی این بار از بغل پاره شده بود و دستمال های داخل آن کثیف بود که برای یک کالای بهداشتی اصلا خوب نیست. فقط نمیدانم دیجی کالا اینو فرستاده یا وقتی ارسال شده پاره شده . ارزش مرجوع کردن هم نداشت و بعد از بیست روز به دستم رسید..	⭐⭐	➖	Not Recommended	0	1	0	0
کیفیت دستمال را کم کردن به همین علت تخفیف میدن اگر خرید قبلی دارید کیفیت جنس دستمال را مقایسه کنید متوجه میشید	⭐	➖	Not Recommended	0	1	0	0
کاش قیمت پایین تر بود .ولی محصولات تنو از کیفیت خوبی برخوردار هستند	⭐⭐⭐⭐	✅	No Opinion	0	0	0	1
بالاتر از قیمت مصرف کننده بود و تولید برج 2 نخرید بیرون ارزون تره	⭐	❌	Not Recommended	0	1	0	0

Evaluation Metrics:

• Precision: 100% (2/2 correct positive predictions)
• Recall: 66.7% (2/3 actual positives caught)
• Specificity: 100% (5/5 correct negative predictions)
• No-Opinion Accuracy: 50% (1 correct out of 2 neutral cases)

Legend:

• ✅ = User recommended (y)
• ❌ = User not recommended (n)
• ➖ = No opinion (-)
• TP = True Positive (True Recommended)
• TN = True Negative
• FP = False Positive
• FN = False Negative

---

🤝 Contribution

Pull requests welcome! For major changes, please open an issue first.

---

Github.com/RezaGooner