Sentiment Analyzer

A full-stack web application that analyzes the sentiment of movie reviews using two approaches:

• Rule-based word analysis
• LLM-based classification via Google's Gemini API

Live Demo: https://sentiment-analyzer-sooty.vercel.app

---

Approach to Sentiment Analysis

This application supports two distinct approaches to sentiment classification:

1. Rule-Based Sentiment Analysis

• A curated list of positive, negative, and neutral words/phrases is used to assess the sentiment of the review.
• The algorithm counts occurrences of these keywords, while also handling edge cases such as:
  • Negations (e.g., "not good" is treated as negative)
  • Multi-word expressions (e.g., "highly recommend" is treated as positive)
  • Neutral expressions (e.g., "okay", "average")
• The final sentiment (Positive, Negative, or Neutral) is determined based on the relative counts.
• An explanation with counts and a computed score is returned to help interpret the result.

Example:

Review: "Absolutely not great. I expected better." Sentiment: Negative Explanation: Rule-based: Score = -2.5 (Pos: 0, Neg: 2, Neutral: 0)

2. LLM-Based Sentiment Analysis (via Gemini)

• Reviews are passed to the Google Gemini API with a prompt instructing it to classify sentiment as Positive, Negative, or Neutral.
• The model's textual response is parsed and matched to one of the three sentiment labels.
• A short explanation (e.g., "Gemini says: Mostly positive with uplifting words") is returned to provide interpretability.
• This method supports more nuanced or complex sentence structures.

---

Features

• Analyze movie reviews using either LLM or rule-based logic
• View past analyses in a searchable, filterable, and sortable table
• Clear explanations provided for both analysis methods
• Delete entries with confirmation dialogs
• Method preference is remembered locally

---

Tech Stack

Frontend

• Next.js (App Router)
• Tailwind CSS for styling
• React Hot Toast for user feedback

Backend

• Express.js using ES Modules
• MongoDB Atlas for database
• Google Gemini API integration
• CORS, dotenv, and body-parser

---

Folder Structure

sentiment-analyzer/
├── backend/                 # Express.js API
│   ├── routes/              # Modular route handlers
│   ├── utils/               # Gemini and rule-based logic
│   ├── models/              # Mongoose schemas
│   └── app.js               # Application entry point
│
├── frontend/                # Next.js application
│   ├── app/                 # App Router pages
│   ├── components/          # Reusable UI components
│   ├── lib/                 # API interaction helpers
│   └── tailwind.config.js   # Tailwind CSS config
│
├── .env                     # Environment variables
└── README.md

---

Local Development

1. Clone the repository

git clone https://github.com/sourabhaprasad/sentiment-analyzer.git
cd sentiment-analyzer

2. Backend Setup

cd backend
npm install
cp .env.example .env
# Fill in MONGO_URI and GEMINI_API_KEY
npm run dev
# Server runs at http://localhost:3001

3. Frontend Setup

cd ../frontend
npm install
cp .env.example .env
# Add NEXT_PUBLIC_API_BASE_URL=http://localhost:3001
npm run dev
# Frontend runs at http://localhost:3000

---

API Endpoints

Method	Endpoint	Description
POST	/analyze	Analyze a movie review
GET	/results	Fetch all analyzed results
GET	/results/:id	Fetch details of a specific result
DELETE	/results/:id	Delete a specific result

---

Extension Patterns (Backend)

• Routes are modular and grouped by resource (/results, /analyze)
• Easy to extend: just add a new file in routes/ and wire it in app.js
• Shared logic like Gemini and rule-based models live in utils/
• Environment and configuration values are centralized in .env

This structure supports quick scalability with minimal boilerplate.

---

Future Enhancements

• Add confidence scores and probability output from LLM
• Build a dashboard to visualize sentiment trends over time
• Support multiple languages for international reviews
• Implement user authentication and history
• Enhance explanation by highlighting detected keywords