URL-Based Topic Classification

Introduction This repository contains the code for simplified experiments based on my seminar paper "Evaluation of Hyperparameter Tuning and Stochastic Gradient Descent for Optimizing URL-Classifying Support Vector Machines" at the University of Passau. These particular experiments are published as part of the "Data Stewardship" course at the Technical University of Vienna. Description The goal of the URL-based topic classification task is to classify websites into different classes, based solely on the URL. These experiments evaluate the impact of hyperparameter tuning on the performance of classification algorithms. Specifically, we evaluate the "alpha" parameter sklearn's "hinge" SGDClassifier, which is a stochastic gradient descent classifier based on a linear support vector machine. We use two datasets for this task: "DMOZ", which labels each URL with one out of 15 categories (e.g. "Arts", "Business", "Science"), and "PhishStorm", which labels each URL as either "phishing" or "benign". Licenses The following licenses are used in this repository: GNU General Public License v3.0 for the code in this repository, found in the experiments directory Creative Commons Attribution 4.0 International for the documentation in this repository, found in the results directory Full Changelog: https://github.com/thomaskirz/ds-url-classification/commits/v1.0.0