SummarAIze — NLP Text Summarizer v3

A full-stack extractive text summarization app built with Node.js + Express (backend) and Angular 17 (frontend), featuring TextRank, TF-IDF, and Word Frequency algorithms, plus Named Entity Recognition and Flesch-Kincaid Readability scoring.

---

Project Structure

text-summarizer/
├── README.md
│
├── backend/
│   ├── package.json
│   └── src/
│       ├── server.js                        # Express app entry point
│       ├── routes/
│       │   └── summarize.routes.js          # POST /api/summarize, /api/summarize/stats
│       ├── controllers/
│       │   └── summarize.controller.js      # Request handlers
│       ├── services/
│       │   └── summarizer.service.js        # TextRank, TF-IDF, Frequency, NER, Readability
│       └── middleware/
│           ├── validateRequest.js           # express-validator middleware
│           └── errorHandler.js             # Global error handler
│
└── frontend/
    ├── angular.json
    ├── tsconfig.json
    ├── tsconfig.app.json
    ├── package.json
    └── src/
        ├── main.ts                          # Bootstrap
        ├── index.html
        ├── styles.scss                      # Global styles
        ├── environments/
        │   └── environment.ts
        └── app/
            ├── app.component.ts
            ├── app.config.ts                # Standalone app config
            ├── app.routes.ts
            ├── models/
            │   └── summarizer.model.ts      # TypeScript interfaces (incl. Entities, Readability)
            ├── services/
            │   └── summarizer.service.ts    # HttpClient API service
            └── components/
                └── summarizer/
                    ├── summarizer.component.ts    # Signals, computed, 5-tab output
                    ├── summarizer.component.html  # Full UI template
                    └── summarizer.component.scss  # Dark terminal styles

---

Getting Started

1. Backend

cd backend
npm install
npm run dev        # nodemon hot reload
# API runs on http://localhost:3000

2. Frontend

cd frontend
npm install
npm start          # Angular dev server
# App runs on http://localhost:4200

---

API Reference

POST /api/summarize

Request body:

{
  "text": "Your long text here...",
  "numSentences": 5,
  "method": "textrank"
}

Field	Type	Required	Description
text	string	✅	Min 100, max 50,000 chars
numSentences	number	❌	1–20 (default: 5)
method	string	❌	textrank, tfidf, or frequency (default: textrank)

Response:

{
  "success": true,
  "data": {
    "summary": "Combined summary text...",
    "summarySentences": ["Sentence 1.", "Sentence 2."],
    "scores": [{ "index": 0, "score": 0.0842 }],
    "stats": {
      "originalWords": 850,
      "originalSentences": 13,
      "summaryWords": 120,
      "summarySentences": 5,
      "compressionRatio": 61.5,
      "readingTimeOriginal": 5,
      "readingTimeSummary": 1
    },
    "keywords": [{ "word": "learning", "score": 4.12 }],
    "entities": {
      "people": ["Steve Jobs"],
      "places": ["California", "United States"],
      "organizations": ["Google", "Microsoft"],
      "numbers": ["2024", "42"]
    },
    "readability": {
      "original": {
        "readingEase": 28.4,
        "easeLabel": "Difficult",
        "gradeLevel": 14.2,
        "avgWordsPerSentence": 22.1,
        "avgSyllablesPerWord": 1.84
      },
      "summary": {
        "readingEase": 31.0,
        "easeLabel": "Difficult",
        "gradeLevel": 13.8,
        "avgWordsPerSentence": 20.4,
        "avgSyllablesPerWord": 1.79
      }
    },
    "method": "textrank"
  }
}

POST /api/summarize/stats

Returns text statistics without summarizing. Includes entities and readability.

---

How the Algorithms Work

TextRank (default — graph-based)

1. Each sentence is a node in a graph
2. Edge weights = cosine similarity between sentence word vectors
3. Similarity matrix is row-normalized
4. Power iteration (PageRank, 30 rounds, damping=0.85) converges on sentence importance scores
5. Top-N sentences selected and reordered to preserve original flow

TF-IDF

1. Each sentence is treated as a document
2. TF-IDF score computed per word per sentence
3. Sentence score = average TF-IDF of its significant words
4. Rare but meaningful words are weighted higher

Word Frequency

1. Word frequency map built from entire text
2. Stop words and numbers removed
3. Frequencies normalized against the max
4. Sentences scored by average normalized frequency of their words
5. Best for news articles with repeated key terms

---

Named Entity Recognition (NER)

Uses the compromise NLP library to extract:

Entity Type	Example
People	Steve Jobs, Elon Musk
Places	California, United States, London
Organizations	Google, Microsoft, OpenAI
Numbers & Values	2024, 42 billion, 15th

---

Readability — Flesch-Kincaid

Two scores computed for both the original text and the summary:

Reading Ease (0–100, higher = easier)

206.835 − (1.015 × avg words/sentence) − (84.6 × avg syllables/word)

Score	Label
90–100	Very Easy
70–89	Easy
60–69	Standard
50–59	Fairly Difficult
30–49	Difficult
0–29	Very Difficult

Grade Level (US school grade)

(0.39 × avg words/sentence) + (11.8 × avg syllables/word) − 15.59

---

Output Tabs

Tab	Content
Summary	Extracted sentences numbered in order, copy button
Keywords	Top 10 TF-IDF keywords with score bar chart
Entities	Color-coded tags — People (purple), Places (green), Orgs (yellow), Numbers (blue)
Readability	Flesch-Kincaid ease meter + grade level for original vs summary
Stats	Word counts, sentence counts, compression ratio, reading time

---

Key Libraries

Library	Purpose
natural	Tokenization, TF-IDF, sentence splitting
compromise	Named entity recognition (people, places, orgs)
stopword	Remove common stop words
express	REST API server
express-validator	Input validation
helmet	HTTP security headers
express-rate-limit	100 req / 15 min rate limiting
@angular/common/http	HttpClient for API calls
@angular/animations	Fade-in & list stagger animations

---

Angular Features Used

• Standalone components (Angular 17, no NgModule)
• Signals (signal(), computed()) for reactive state
• HttpClient with typed Observable responses
• Animations (@fadeIn, @listStagger)
• FormsModule with ngModel two-way binding
• RxJS catchError, map for error handling