DocumentClassifier

Overview

DocumentClassifier is a Swift framework for classifying documents into one of five categories (Business, Entertainment, Politics, Sports, and Technology). It uses a CoreML model trained with 1,500 news articles from the BBC.

Features

iOS 11.0+, macOS 10.13+, tvOS 11.0+, watchOS 4.0+
100% Test Coverage
Best CV Score: 0.965333333333

Usage

Swift 4 Released (Sample Article)

let text = articleText
guard let classification = classifier.classify(text) else { return }
print(classification.prediction) // Technology: 0.42115752953489294
print(classification.allResults) // Business: 0.141, Entertainment: 0.138, Politics: 0.113, Sports: 0.187, Technology: 0.421

Installation

CocoaPods

CocoaPods is a centralized dependency manager for Cocoa projects. To install DocumentClassifier with CocoaPods:

Make sure the latest version of CocoaPods is installed.
Add DocumentClassifier to your Podfile:

use_frameworks!

pod 'DocumentClassifier', '1.2.0'

Run pod install.

Example App

NewsClassifier is an example app using the framework.

Model

Model Link
Best CV Score: 0.965333333333
Trained using 1,500 news articles from the BBC from 2004-2005 (see references)
Converted from scikit-learn Pipeline using coremltools.
Based on the LinearSVC classifier.

Author

Todd Kramer, todd@tekramer.com

References

BBC Datasets
D. Greene and P. Cunningham. "Practical Solutions to the Problem of Diagonal Dominance in Kernel Document Clustering", Proc. ICML 2006. link
Vadym Markov, SentimentPolarity
Awesome Core ML Models
scikit-learn
Apple Machine Learning
CoreML Framework
coremltools

toddkramer/DocumentClassifier