AgenticAI

This repository contains an implementation of AgenticAI, an AI-powered system designed to perform sentiment analysis on app reviews from platforms like the Google Play Store. The AI agent utilizes state-of-the-art models like Hugging Face, OpenAI, and other custom models for improved analysis.

Features

• Sentiment Analysis: Classify reviews as positive, negative, or neutral based on the text and emojis.
• Emoji-Aware Analysis: Preprocesses emoji symbols to better understand their emotional context.
• Integrates Hugging Face & OpenAI: Leverages popular AI platforms for advanced natural language processing.
• Customizable: Easily integrate new models or preprocessors for specific use cases.

Prerequisites

Before running the project, ensure you have the following:

• Node Js 18 >: The project is built using JS.
• LangFlow: Used to create AI agents using a flow-based programming approach.
• Hugging Face API Key: To access pre-trained models from Hugging Face.
• OpenAI API Key: For advanced sentiment interpretation and analysis.
• MongoDB (optional): For storing results.
• Google Play Scraper: Used for fetching app reviews from Google Play Store.

Setup

1. Clone the Repository

git clone https://github.com/shivrajkhetri7/agenticai.git
cd AgenticAI

2. Install Dependencies

Create a virtual environment and install the required libraries.

npm install

3. API Keys

Ensure that you have Hugging Face and OpenAI API keys.

• For Hugging Face: Create an account on Hugging Face and get your API key.
• For OpenAI: Create an account on OpenAI and get your API key.

4. Configuration

Update the env file with your API keys and other configurations:

HUGGINGFACE_API_KEY = "your-huggingface-api-key"
OPENAI_API_KEY = "your-openai-api-key"
MONGO_URI = "mongodb://localhost:27017/"

5. Running the Application

Start the agent using LangFlow or directly via scripts.

node analyzeSentiment.js

Enter Enter Your Application Id:

Usage

1. Input Data

You can input app details (e.g., app ID) to fetch reviews, or use a file containing the raw reviews.

2. Review Processing

The agent processes reviews by:

• Scraping the reviews from Google Play Store using the app ID.
• Preprocessing the reviews (emoji to text conversion, cleaning).
• Performing sentiment analysis using Hugging Face, OpenAI, and custom models.
• Storing the sentiment results in MongoDB.

3. Output

The agent returns sentiment analysis for each review. The output includes:

• Sentiment (Positive/Negative/Neutral)
• Confidence scores
• Emoji-based sentiment scores (if applicable)

4. Example Output

{
  "review": "😍😇😇🙏🙏❤️",
  "sentiment": "Positive",
  "score": 0.9,
  "emoji_sentiment": "Very Positive"
}

Customization

You can easily add new models or alter the flow using LangFlow's interface. Some potential customizations include:

• Model Selection: Switch out Hugging Face and OpenAI models for custom-trained models.
• Text Preprocessing: Adjust the preprocessing pipeline to remove specific words, symbols, etc.
• Emoji Handling: Fine-tune the emoji-to-text conversion for better accuracy.

Contributing

We welcome contributions to enhance the features of AgenticAI. Here’s how you can help:

Explanation of Sections:

• Project Overview: Describes the purpose of the repository and the core features.
• Prerequisites: Lists software and tools required to run the project.
• Setup: Provides step-by-step instructions on how to set up the project locally.
• Usage: Explains how to run the project and how to use the input data (app ID or raw reviews) for sentiment analysis.
• Customization: Offers insights into how users can tweak the agent to better suit their needs.
• Contributing: Provides guidelines on how others can contribute to the project.
• License: Specifies the project's licensing (MIT in this case).

This README.md should guide you and other developers through the process of setting up and using the AgenticAI repository. Let me know if you need any more details added!