Classifying and labeling the relationships between cities with high levels of co-occurrence on the English Wikipedia

Abstract

This study delves into three different approaches of document classification in order to successfully classify the type of relationship between European cities: ”LDA Topic modeling, Word embedding classification and Word frequency representation”. The first method provides a distribution of topics, the second provides hard classification, while the last method uses word frequency metrics to represent a document by its most relevant words. LDA topic modeling and word embedding classification provided very similar results for a dataset of 311.000 paragraphs, indicating a serious level of accuracy, and proving that they could both be used for classification.