Text Summarization and Keyword Extraction for Swift

Features • Requirements • Installation • Usage • Documentation • Contributing

Reductio is a high-performance Swift library that implements the TextRank algorithm for automatic text summarization and keyword extraction. Built with Swift 6's strict concurrency in mind, it provides a modern, safe, and efficient way to analyze and extract meaning from text.

TextRank is an unsupervised graph-based ranking algorithm inspired by Google's PageRank. It builds a graph representation of text where vertices are words or sentences, and edges represent semantic relationships. Through iterative cal 8000 culation, it identifies the most important elements in the text.

• 📰 News Summarization: Extract key points from articles
• 🔍 SEO Optimization: Identify important keywords for content
• 📚 Academic Research: Summarize research papers and extract key concepts
• 📱 Social Media: Analyze and summarize user-generated content
• 🏢 Business Intelligence: Process reports and extract insights
• 💬 Chatbots: Generate concise responses from knowledge bases

• 🔑 Keyword Extraction: Identify the most relevant keywords using graph-based ranking
• 📝 Text Summarization: Extract key sentences while preserving context
• 🌐 Language Support: Optimized for English with extensible architecture
• ⚡ High Performance: Efficient processing of documents up to 10,000+ words
• 🔒 Thread-Safe: Full Swift 6 strict concurrency compliance
• 📦 Zero Dependencies: Pure Swift implementation

• Modern Swift APIs: Native async/await support
• Value Semantics: Immutable structs for Keyword and Summarizer
• NaturalLanguage Framework: Leverages Apple's ML-powered text processing
• Flexible Configuration: Customizable compression ratios and result counts
• Extension-Friendly: Convenient String extensions for quick access

• Swift: 6.0+
• Xcode: 15.0+
• Platforms:
  • iOS 13.0+
  • macOS 12.0+
  • tvOS 13.0+
  • watchOS 6.0+

Add Reductio to your Package.swift dependencies:

dependencies: [
  .package(url: "https://github.com/fdzsergio/Reductio.git", from: "1.6.0")
]

Then add it to your target:

targets: [
  .target(
    name: "YourApp",
    dependencies: ["Reductio"]
  )
]

import Reductio

let text = """
Apple Inc. announced groundbreaking updates to its developer tools at WWDC 2024. 
The company introduced Swift 6 with major improvements to concurrency and performance. 
Developers praised the new features, particularly the enhanced type safety and 
async/await improvements that make concurrent programming more intuitive.
"""

// Extract top keywords
let keywords = await Reductio.keywords(from: text, count: 5)
print(keywords)  // ["swift", "developers", "improvements", "concurrency", "features"]

// Get a summary
let summary = await Reductio.summarize(text: text, count: 2)
summary.forEach { print("• \($0)") }

// Extract all keywords ranked by importance
let allKeywords = await Reductio.keywords(from: text)

// Get top 10 keywords
let topKeywords = await Reductio.keywords(from: text, count: 10)

// Extract keywords with 70% compression (top 30% of keywords)
let compressedKeywords = await Reductio.keywords(from: text, compression: 0.7)

// Synchronous keyword extraction
let keywords = text.keywords

// Custom configuration
let topKeywords = text.keywords(count: 5)

// Get all sentences ranked by importance
let rankedSentences = await Reductio.summarize(text: text)

// Extract top 3 most important sentences
let summary = await Reductio.summarize(text: text, count: 3)

// Summarize with 80% compression (keep 20% of sentences)
let conciseSummary = await Reductio.summarize(text: text, compression: 0.8)

// Quick summarization
let summary = text.summarize

// Custom length
let shortSummary = text.summarize(count: 2)

struct DocumentAnalyzer {
  static func analyze(_ document: String) async -> DocumentInsights {
    async let keywords = Reductio.keywords(from: document, count: 10)
    async let summary = Reductio.summarize(text: document, count: 5)
    
    return await DocumentInsights(
      keywords: keywords,
      summary: summary,
      readingTime: estimateReadingTime(document)
    )
  }
}

extension String {
  func seoAnalysis() async -> SEOReport {
    let keywords = await Reductio.keywords(from: self, count: 20)
    let keywordDensity = calculateDensity(keywords: keywords, in: self)
    
    return SEOReport(
      primaryKeywords: Array(keywords.prefix(5)),
      secondaryKeywords: Array(keywords.dropFir
8000
st(5)),
      keywordDensity: keywordDensity,
      suggestedMetaDescription: self.summarize(count: 1).first ?? ""
    )
  }
}

Get started with documentation at https://fdzsergio.github.io/Reductio/

1. Text Preprocessing

   • Sentence segmentation using NaturalLanguage framework
   • Word tokenization and normalization
   • Stopword removal (common words like "the", "is", "at")
   • Lemmatization to reduce words to base forms

2. Graph Construction

   • Each word/sentence becomes a vertex
   • Edges connect co-occurring elements
   • Edge weights represent semantic similarity

3. Iterative Ranking

   • Initial scores assigned to all vertices
   • Scores propagate through edges
   • Convergence achieved after ~30 iterations

4. Result Extraction

   • Vertices ranked by final scores
   • Top elements returned as keywords/summary

For optimal performance:

• Process documents under 10,000 words
• Use compression ratios between 0.7-0.9
• Cache results for repeated analysis
• Consider chunking very large documents

We welcome contributions! Please see our Contributing Guide for details.

# Clone the repository
git clone https://github.com/fdzsergio/Reductio.git
cd Reductio

# Build the project
swift build

# Run tests
swift test

# Generate documentation
swift package plugin generate-documentation

• Follow Swift API Design Guidelines
• Use swift-format for consistency
• Write tests for new features
• Update documentation as needed

Reductio is released under the MIT License. See LICENSE for details.

• 🎨 Logo design by @cristinareina
• 📚 Based on TextRank by Rada Mihalcea and Paul Tarau
• 🙏 Thanks to all contributors

Sergio Fernández
📧 fdz.sergio@gmail.com
🐦 @fdzsergio
💼 LinkedIn

Made with ❤️ and Swift