Arabic Sentiment Analysis API

This project provides a Flask-based REST API for analyzing sentiment in Arabic text. It can analyze individual messages, batch texts, and entire conversations, providing detailed sentiment analysis including negativity percentages.

Features

• Single text sentiment analysis
• Batch text analysis
• Conversation analysis with role-based sentiment tracking
• Negativity percentage calculation
• Confidence scores for predictions

Prerequisites

• Python 3.7 or higher
• pip (Python package manager)
• Fine-tuned model files (see Model Files section below)

Model Files

⚠️ Important: The model files are not included in this repository due to size limitations. You can:

1. Run the following command to generate the model files:

python fine_tune.py

2. Ensure the following files are present in the directory:
   • tokenizer.json
   • vocab.txt
   • pytorch_model.bin
   • config.json

Installation

1. Clone the repository:

git clone [repository-url]
cd [repository-name]

2. Download and place the model files as described in the Model Files section.

3. Install the required dependencies:

pip install -r requirements.txt

Running the Application

Start the server by running:

python app.py

The API will be available at http://localhost:5000

API Endpoints

1. Single Text Analysis

Analyze the sentiment of a single text.

Endpoint: POST /analyze

Request Body:

{
    "text": "هذا المنتج رائع جداً"
}

Response:

{
    "text": "هذا المنتج رائع جداً",
    "sentiment": "positive",
    "confidence": "95.5%"
}

2. Batch Analysis

Analyze multiple texts at once.

Endpoint: POST /analyze_batch

Request Body:

{
    "texts": [
        "هذا المنتج رائع جداً",
        "لم يعجبني هذا المنتج على الإطلاق"
    ]
}

3. Conversation Analysis

Analyze a complete conversation with role-based sentiment tracking.

Endpoint: POST /analyze_conversation

Request Body:

{
    "conversation": "Customer: المنتج الذي استلمته لا يعمل بشكل صحيح\nAgent: عذراً جداً على هذه التجربة\nCustomer: حسناً، شكراً لك"
}

Response:

{
    "overall_sentiment": "negative",
    "negative_messages": [
        {
            "message": "المنتج الذي استلمته لا يعمل بشكل صحيح",
            "confidence": "85.5%"
        }
    ],
    "sentiment_distribution": {
        "positive": 0,
        "neutral": 2,
        "negative": 1
    },
    "negativity_percentage": "33.3%"
}

Response Fields

• sentiment: Can be "positive", "neutral", or "negative"
• confidence: Confidence score of the prediction
• overall_sentiment: Overall conversation sentiment
• negative_messages: List of detected negative messages
• sentiment_distribution: Count of different sentiments in the conversation
• negativity_percentage: Percentage of negative messages in the conversation

Error Handling

The API returns appropriate HTTP status codes:

• 200: Successful request
• 400: Bad request (missing or invalid parameters)
• 500: Server error

Development

The application is built using:

• Flask for the web server
• Transformers library for the sentiment analysis model
• PyTorch as the deep learning backend