How can you perform sentiment analysis using C# and ML.NET?

You can perform sentiment analysis in C# using ML.NET by training or consuming a machine learning model that classifies text as positive or negative.

A common workflow is:

• Install ML.NET packages
• Prepare training data
• Build and train a pipeline
• Evaluate the model
• Predict sentiment on new text

Official docs: ML.NET Documentation

1. Create a Console Project

dotnet new console -n SentimentAnalysisApp
cd SentimentAnalysisApp

Install the ML.NET package:

dotnet add package Microsoft.ML

ML.NET NuGet package:
Microsoft.ML NuGet Package

2. Prepare Training Data

Create a file named sentiment-data.tsv.

Example:

Sentiment	Text
1	I love this product
1	This is amazing
0	I hate this item
0	This is terrible

• 1 = Positive
• 0 = Negative

3. Create Data Models

Create a file called SentimentData.cs.

using Microsoft.ML.Data;

// Input data class
public class SentimentData
{
    // First column from dataset
    [LoadColumn(0)]
    public bool Sentiment { get; set; }

    // Second column from dataset
    [LoadColumn(1)]
    public string Text { get; set; }
}

// Prediction output class
public class SentimentPrediction
{
    // Predicted sentiment
    [ColumnName("PredictedLabel")]
    public bool Prediction { get; set; }

    // Confidence score
    public float Probability { get; set; }

    // Raw model score
    public float Score { get; set; }
}

4. Train the Model

Replace Program.cs with:

using Microsoft.ML;
using Microsoft.ML.Data;

// Create ML context
var mlContext = new MLContext();

// Load training data
IDataView dataView = mlContext.Data.LoadFromTextFile<SentimentData>(
    path: "sentiment-data.tsv",
    hasHeader: true
);

// Split dataset into train/test sets
var splitData = mlContext.Data.TrainTestSplit(
    dataView,
    testFraction: 0.2
);

// Build ML pipeline
var pipeline = mlContext.Transforms.Text.FeaturizeText(
        outputColumnName: "Features",
        inputColumnName: nameof(SentimentData.Text)
    )
    .Append(
        mlContext.BinaryClassification.Trainers.SdcaLogisticRegression(
            labelColumnName: nameof(SentimentData.Sentiment),
            featureColumnName: "Features"
        )
    );

// Train model
var model = pipeline.Fit(splitData.TrainSet);

// Evaluate model
var predictions = model.Transform(splitData.TestSet);

var metrics = mlContext.BinaryClassification.Evaluate(
    predictions,
    labelColumnName: nameof(SentimentData.Sentiment)
);

// Print evaluation metrics
Console.WriteLine($"Accuracy: {metrics.Accuracy:P2}");
Console.WriteLine($"F1 Score: {metrics.F1Score:P2}");
Console.WriteLine($"AUC: {metrics.AreaUnderRocCurve:P2}");

// Create prediction engine
var predictionEngine = mlContext.Model.CreatePredictionEngine
    <SentimentData, SentimentPrediction>(model);

// Example prediction
var sample = new SentimentData
{
    Text = "This app is fantastic!"
};

// Predict sentiment
var prediction = predictionEngine.Predict(sample);

// Display result
Console.WriteLine($"Text: {sample.Text}");
Console.WriteLine($"Prediction: {(prediction.Prediction ? "Positive" : "Negative")}");
Console.WriteLine($"Probability: {prediction.Probability:P2}");

5. Run the Application

dotnet run

Example output:

Accuracy: 95.00%
F1 Score: 94.00%
AUC: 97.00%

Text: This app is fantastic!
Prediction: Positive
Probability: 98.12%

How ML.NET Processes Text

The important step is:

FeaturizeText()

This converts raw text into numerical vectors using NLP techniques such as:

• Tokenization
• N-grams
• TF-IDF weighting
• Text normalization

The classifier then learns patterns associated with positive or negative sentiment.

Using a Pretrained Model

Instead of training from scratch, you can also use:

• ONNX models
• Azure AI services

Hugging Face transformer models exported to ONNX

ONNX with ML.NET:
ML.NET ONNX Guide

Real-World Use Cases

Sentiment analysis is commonly used for:

• Customer review analysis
• Social media monitoring
• Chat moderation
• Support ticket prioritization
• Product feedback analytics

Improving Accuracy

For production systems:

• Use larger datasets
• Clean noisy text
• Remove spam/duplicates
• Use transformer-based models
• Balance positive/negative samples
• Add multilingual support if needed

Useful ML.NET Resources

ML.NET Official Site

ML.NET Tutorials

ML.NET GitHub Repository