Skip to content

CHANGELOG.md

Latest commit

 

History

History
143 lines (98 loc) · 4.5 KB

CHANGELOG.md

File metadata and controls

143 lines (98 loc) · 4.5 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[3.0.0] - 2026-01-27

🎉 Major Release: Multi-Engine Architecture

This release transforms SentimentAnalyzer from a single ML.NET-based package into a modular, multi-engine sentiment analysis suite with 4 distinct NuGet packages.

New Packages

SentimentAnalyzer.Core 3.0.0

  • New Package: Lightweight VADER-based sentiment analysis (<1 MB)
  • Uses VaderSharp2 for rule-based sentiment scoring
  • ~75% accuracy on general sentiment tasks
  • No ML model files required
  • Perfect for size-constrained applications

SentimentAnalyzer.Onnx 3.0.0

  • New Package: High-accuracy TinyBERT transformer model (~18 MB)
  • 94.2% accuracy on sentiment classification
  • Embedded ONNX model (no external downloads)
  • Supports .NET Standard 2.0, .NET 8, and .NET 10
  • Cross-platform (Windows, Linux, macOS, iOS, Android)

SentimentAnalyzer.Onnx.Multilingual 3.0.0

  • New Package: Multilingual DistilBERT model supporting 104 languages
  • ~88% accuracy across multiple languages
  • Downloads 270 MB model on first use (cached locally)
  • Async download support with progress reporting
  • Fixed HuggingFace ONNX export bug (correctly exported model hosted on GitHub Releases)

SentimentAnalyzer 3.0.0 (Legacy)

  • Maintained for backward compatibility
  • Upgraded ML.NET from 1.5.5 to 5.0.0
  • No API changes - drop-in replacement for v2.x
  • Multi-targeting: .NET Standard 2.0, .NET 8, .NET 10

New Features

  • Fluent Builder API: SentimentAnalyzer.CreateBuilder().UseVader().Build()
  • ISentimentEngine Interface: Create custom sentiment engines
  • Thread-Safe: All engines are thread-safe and can be reused
  • Comprehensive Test Suite: 58 tests across Core.Tests and integration tests
  • Sample Applications: Console, Blazor WASM, MAUI, BenchmarkDotNet samples

Architecture Changes

  • Modular design allowing mix-and-match of engines
  • Shared core abstractions (ISentimentEngine, SentimentResult, SentimentLabel)
  • Package size optimization (choose what you need)

Documentation

  • Comprehensive README with engine comparison table
  • Performance benchmarks and accuracy metrics
  • Migration guide for v2.x users
  • API reference for all packages

Breaking Changes from 2.x

  • New namespace structure (see migration guide)
  • Builder pattern required for new engines
  • Legacy Sentiments.Predict() API moved to SentimentAnalyzer package

Model Hosting

  • TinyBERT model: Embedded in SentimentAnalyzer.Onnx package
  • DistilBERT model: Hosted on GitHub Releases (correctly exported ONNX model)
  • Vocab files: Downloaded from HuggingFace

[2.0.0] - 2026-01-22

Added

  • Multi-targeting support for .NET Standard 2.0, .NET 8, and .NET 10
  • Package README included in NuGet package
  • Additional package tags for better discoverability

Changed

  • Upgraded ML.NET from 1.5.5 to 5.0.0 - This is a major dependency upgrade
  • Updated package description and metadata
  • Test project upgraded to .NET 8 with latest test framework versions

Compatibility

This release maintains full backward compatibility with v1.x:

  • The API (Sentiments.Predict(text)) remains unchanged
  • The SentimentPrediction model is unchanged
  • Existing code will work without modifications

Migration Guide

No code changes required. Simply update your package reference:

<!-- Before -->
<PackageReference Include="SentimentAnalyzer" Version="1.2.3" />

<!-- After -->
<PackageReference Include="SentimentAnalyzer" Version="2.0.0" />

Platform Support

Platform Supported
.NET Framework 4.6.1+ ✅ (via netstandard2.0)
.NET Core 2.0+ ✅ (via netstandard2.0)
.NET 5/6/7 ✅ (via netstandard2.0)
.NET 8 (LTS) ✅ Native
.NET 10 (LTS) ✅ Native
.NET MAUI ✅
Blazor (Server/WASM) ✅
Unity ✅ (via netstandard2.0)

[1.2.3] - 2021-xx-xx

Changed

  • Performance improvements using singleton pattern
  • Based on ML.NET version 1.5.5

[1.0.0] - 2020-xx-xx

Added

  • Initial release
  • Binary sentiment analysis (positive/negative)
  • Offline, on-device prediction
  • Thread-safe singleton pattern