Text Classifier

This package provides a C# wrapper for the library: https://github.com/lytics/multibayes - providing multiclass naive Bayesian document classification in Unity

A basic example looks like this

using UnityEngine;
using SkybirdGames.TextClassifier;

public class TextClassiferTest : MonoBehaviour
{
    [SerializeField] private NaiveBayesTextClassifierData trainingData;

    void Start()
    {
        var textClassifier = new NaiveBayesTextClassifier(trainingData.ToJson());

        print(textClassifier.Predict(text: "Hello there my friend", threshold: 0.4));

        // result = "greetings"
    }
}

Installation

To install the package in your Unity project:

1. Open the package manager
2. Press the "+" button
3. Select "add package from git URL"
4. Enter: https://github.com/claytonpeterson/unity-text-classifier.git
5. Press "add"

Training Data Set

The underlying library requires a json training string in the format

[
    {
        "Text": "Example statement",
        "Classes": [ "class" ]
    },
    {
        "Text": "Example statement",
        "Classes": [ "class" ]
    }...
]

About the format

1. Text contains a statement that provides an example of the class
   "Text": "Hello my friend"
2. Classes contains the collection of classes that the statement belongs to
   "Classes": [ "greetings", "friendship" ]

NaiveBayesTextClassifierData

You can write your json training data manually, or to make things easier you can create a NaiveBayesTextClassifierData scriptable object in Unity

How to Create

1. To create the training object select "Assets/Create/SkybirdGames/Naive Bayes Text Classifier/New Dataset"
2. Next, select the object in the inspector, open it's TrainingData array and press the "+" button to add a new class
3. Finally, open the class's Texts array, and press the "+" button to add example statements to the class

NaiveBayesTextClassifier

The NaiveBayesTextClassifier class is initialized with a string containing the json training data

var textClassifier = new NaiveBayesTextClassifier(jsonTrainingData);

The Predict method returns the predicted class
The Threshold is used to filter out results that have a lower probability than the threshold

var text = "Hide behind that rock!"
var threshold = 0.4;
var result = textClassifier.Predict(text, threshold);

// result = "cover"

Included Samples

The folder "Text Classifier/Samples/TrainingData" includes:

1. A raw json data file ("combat_training_data") that is trained to recognize "combat dialogue"
2. A NaiveBayesTextClassifierData object that is trained to recognize greetings and farewells