📖 Sentiment Analyzer

A reusable sentiment analysis library for Node.js / TypeScript that works with Gemini (Google Generative Language API) and OpenAI (Chat Completions API).

It accepts single or multiple reviews and returns:

• Overall sentiment (positive / negative / neutral)
• Aspects detected in the text (e.g., battery, camera, support)
• Aspect-based sentiment (per aspect, whether positive/negative/neutral)
• Aggregated summary across multiple reviews

🚀 Installation

npm i @rathakrishna/sentiment-analysis

🔑 Setup

You’ll need an API key from either:

• Gemini (Google Generative Language API) → Get a key
• OpenAI → Get a key

Save it in your environment (.env file or process env vars):

# For Gemini
GEMINI_API_KEY=your_gemini_api_key

# For OpenAI
OPENAI_API_KEY=your_openai_api_key

📦 Usage

Import

import { GeminiSentimentAnalyzer, OpenAISentimentAnalyzer } from "@rathakrishna/sentiment-analyzer";

1. Gemini Example

const analyzer = new GeminiSentimentAnalyzer(process.env.GEMINI_API_KEY!);

const result = await analyzer.analyzeReviews([
  "The battery life is amazing, but the camera is disappointing.",
  "Support was fast and helpful."
]);

console.log(JSON.stringify(result, null, 2));

2. OpenAI Example

const analyzer = new OpenAISentimentAnalyzer(process.env.OPENAI_API_KEY!);

const result = await analyzer.analyzeReviews("The screen is bright but the speakers are weak.");

console.log(JSON.stringify(result, null, 2));

📥 Input Format

You can pass either:

• Single review (string)

await analyzer.analyzeReviews("Battery lasts all day but camera is bad.");

• Multiple reviews (string[])

await analyzer.analyzeReviews([
  "Battery lasts all day!",
  "Camera is terrible at night.",
  "Support was helpful."
]);

📤 Output Format

The library always returns:

{
  reviews: [
    {
      text: string;                 // Original review
      sentiment: "positive" | "negative" | "neutral";
      aspects: string[];             // Detected aspects
      aspect_sentiments: Array<{
        aspect: string;
        sentiment: "positive" | "negative" | "neutral";
      }>;
    },
    ...
  ],
  aggregated: {
    overall_sentiment: "positive" | "negative" | "neutral";
    aspect_summary: {
      [aspect: string]: {
        positive: number;
        negative: number;
        neutral: number;
      }
    }
  }
}

✅ Example Response

Input:

{
  "reviews": [
    "Battery lasts all day!",
    "Camera is terrible at night.",
    "Support was helpful."
  ]
}

Output:

{
  "reviews": [
    {
      "text": "Battery lasts all day!",
      "sentiment": "positive",
      "aspects": ["battery"],
      "aspect_sentiments": [
        { "aspect": "battery", "sentiment": "positive" }
      ]
    },
    {
      "text": "Camera is terrible at night.",
      "sentiment": "negative",
      "aspects": ["camera", "low light"],
      "aspect_sentiments": [
        { "aspect": "camera", "sentiment": "negative" },
        { "aspect": "low light", "sentiment": "negative" }
      ]
    },
    {
      "text": "Support was helpful.",
      "sentiment": "positive",
      "aspects": ["support"],
      "aspect_sentiments": [
        { "aspect": "support", "sentiment": "positive" }
      ]
    }
  ],
  "aggregated": {
    "overall_sentiment": "positive",
    "aspect_summary": {
      "battery": { "positive": 1, "negative": 0, "neutral": 0 },
      "camera": { "positive": 0, "negative": 1, "neutral": 0 },
      "low light": { "positive": 0, "negative": 1, "neutral": 0 },
      "support": { "positive": 1, "negative": 0, "neutral": 0 }
    }
  }
}

📚 API

analyzeReviews(reviews: string | string[]): Promise<AnalysisResult>

• Input: Single string or array of review strings
• Output: AnalysisResult

interface AspectSentiment {
  aspect: string;
  sentiment: "positive" | "negative" | "neutral";
}

interface ReviewAnalysis {
  text: string;
  sentiment: "positive" | "negative" | "neutral";
  aspects: string[];
  aspect_sentiments: AspectSentiment[];
}

interface AggregatedAnalysis {
  overall_sentiment: "positive" | "negative" | "neutral";
  aspect_summary: Record<string, { positive: number; negative: number; neutral: number }>;
}

interface AnalysisResult {
  reviews: ReviewAnalysis[];
  aggregated: AggregatedAnalysis;
}

⚡ Features

• Works with both Gemini & OpenAI
• Supports single or multiple reviews
• Returns overall + aspect-based sentiment
• TypeScript-ready with .d.ts typings
• Can be used in Node.js backends, serverless functions, or APIs

🛡 License

MIT © 2025 Rathakrishnan