SentimentAnalyzer.Core 3.0.0

SentimentAnalyzer

On-device (offline) Sentiment Analysis for .NET applications

A cross-platform, privacy-first library for sentiment analysis that runs entirely on-device without requiring internet connectivity.

✨ Features

• 🔒 Privacy-first - All processing happens on-device
• 📴 Fully offline - No internet connection required
• 🚀 Cross-platform - .NET Standard 2.0, .NET 8, .NET 10, MAUI, Blazor
• ⚡ Multiple engines - Choose between speed, accuracy, or both
• 🌍 Multilingual - Support for 104 languages

🚀 Quick Start

Option A: Lightweight (VADER) - Recommended for most apps

dotnet add package SentimentAnalyzer.Core

using SentimentAnalyzer;

// Simple usage
var analyzer = new SentimentAnalyzer();
var result = analyzer.Analyze("I love this product!");

Console.WriteLine($"Sentiment: {result.Label}");     // Positive
Console.WriteLine($"Score: {result.Score:F2}");      // 0.87
Console.WriteLine($"Confidence: {result.Confidence:F2}");  // 0.74

Option B: High Accuracy (TinyBERT) - English Only

dotnet add package SentimentAnalyzer.Core
dotnet add package SentimentAnalyzer.Onnx

using SentimentAnalyzer;
using SentimentAnalyzer.Onnx;

// Use TinyBERT for higher accuracy (embedded model, ~18MB)
var analyzer = SentimentAnalyzer.CreateBuilder()
    .UseTinyBert()
    .Build();

var result = analyzer.Analyze("This product is amazing!");
Console.WriteLine($"Sentiment: {result.Label}");  // Positive
Console.WriteLine($"Confidence: {result.Confidence:P0}");  // 99%

Option C: Multilingual (DistilBERT) - 104 Languages

dotnet add package SentimentAnalyzer.Core
dotnet add package SentimentAnalyzer.Onnx.Multilingual

using SentimentAnalyzer;
using SentimentAnalyzer.Onnx.Multilingual;

// Use DistilBERT for multilingual support (downloads ~270MB model on first use)
var analyzer = SentimentAnalyzer.CreateBuilder()
    .UseDistilBert()
    .Build();

// Supports 104 languages!
var result = analyzer.Analyze("Ce produit est fantastique!");  // French
Console.WriteLine($"Sentiment: {result.Label}");  // Positive

// Also works with: Arabic, Chinese, German, Spanish, Hindi, Japanese, etc.

📥 Note on DistilBERT Multilingual: The ~270MB model is automatically downloaded from GitHub Releases on first use and cached locally at %LOCALAPPDATA%\SentimentAnalyzer\models\. Subsequent uses are fully offline. See full list of 104 supported languages.

Option D: Legacy API (Backward Compatible)

dotnet add package SentimentAnalyzer

using SentimentAnalyzer;

// v1.x/v2.x API still works
var result = Sentiments.Predict("This product is amazing!");
Console.WriteLine($"Positive: {result.Prediction}");  // true

� Choosing the Right Package

Package Comparison

Speed ratings: ⚡⚡⚡⚡⚡ (fastest) to ⚡ (slowest)

Platform Support

Note: ONNX models require native C++ libraries and cannot run in Blazor WebAssembly. Use Core (VADER) for WASM or Blazor Server for ML models.

Use Cases

�📖 API Reference

Core Package

SentimentAnalyzer Class

// Default constructor (uses VADER engine)
var analyzer = new SentimentAnalyzer();

// With options
var analyzer = new SentimentAnalyzer(SentimentAnalyzerOptions.Binary);

// Builder pattern
var analyzer = SentimentAnalyzer.CreateBuilder()
    .WithNeutralThresholds(0.3, 0.7)  // Customize neutral zone
    .Build();

SentimentResult

SentimentAnalyzerOptions

// Presets
SentimentAnalyzerOptions.Default   // Neutral zone: 0.4-0.6
SentimentAnalyzerOptions.Binary    // No neutral (positive/negative only)
SentimentAnalyzerOptions.Strict    // Narrow neutral: 0.45-0.55
SentimentAnalyzerOptions.Lenient   // Wide neutral: 0.3-0.7

Batch Processing

var analyzer = new SentimentAnalyzer();

var reviews = new[] {
    "Great product!",
    "Terrible service.",
    "It's okay I guess."
};

var results = analyzer.AnalyzeBatch(reviews);
foreach (var result in results)
{
    Console.WriteLine($"{result.Label}: {result.Text}");
}

🛠️ Advanced Usage

Custom Engine

// Implement ISentimentEngine for custom engines
public class MyCustomEngine : ISentimentEngine
{
    public string Name => "Custom";

    public SentimentResult Analyze(string text) { /* ... */ }
    public IReadOnlyList<SentimentResult> AnalyzeBatch(IEnumerable<string> texts) { /* ... */ }
    public void Dispose() { }
}

// Use with builder
var analyzer = SentimentAnalyzer.CreateBuilder()
    .WithEngine(_ => new MyCustomEngine())
    .Build();
```

### TinyBERT with Custom Options

```csharp
using SentimentAnalyzer.Onnx;
using Microsoft.ML.OnnxRuntime;

// Configure ONNX Runtime session options
var sessionOptions = new SessionOptions
{
    GraphOptimizationLevel = GraphOptimizationLevel.ORT_ENABLE_ALL,
    ExecutionMode = ExecutionMode.ORT_PARALLEL
};

var analyzer = SentimentAnalyzer.CreateBuilder()
    .UseTinyBert(sessionOptions)
    .WithBinaryClassification()  // Disable neutral classification
    .Build();

DistilBERT Multilingual with Progress

using SentimentAnalyzer.Onnx.Multilingual;

var progress = new Progress<double>(p =>
    Console.WriteLine($"Downloading model: {p:P0}"));

var analyzer = await SentimentAnalyzer.CreateBuilder()
    .UseDistilBertMultilingualAsync(progress: progress);

// Model is now cached locally, subsequent uses are offline

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📚 Learn More

• Blog Post: On-device Machine Learning with SentimentAnalyzer
• CHANGELOG for version history
• GitHub Releases for model downloads

⚡ Performance Comparison

Speed ratings: ⚡⚡⚡⚡⚡ (fastest) to ⚡ (slowest). Accuracy tested on various datasets. � Migration from v2.x

If using Sentiments.Predict():

No changes needed! The legacy API continues to work:

<PackageReference Include="SentimentAnalyzer" Version="3.0.0" />

To use new engines (smaller, faster, or more accurate):

// Before (ML.NET, ~15MB)
var result = Sentiments.Predict("I love this!");

// After (VADER, <1MB - fastest)
var analyzer = new SentimentAnalyzer();
var result = analyzer.Analyze("I love this!");

// Result now has more properties:
// result.Label (enum), result.Confidence, result.IsPositive, etc.

📚 Learn More

• Blog Post: On-device Machine Learning with SentimentAnalyzer
• Sample Applications - Console, Blazor, MAUI examples
• CHANGELOG - Version history and migration guides
• GitHub Releases - Model downloads

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.