Korean football stylometry by using the football events networks

Abstract

Web-casting text is a text broadcast service for sports games. Just by using the web-casting text, most of the content of the game is delivered. In this study, we make a word network by using 389,139 football game events from 344 matches of Korean national team and analyze the network through the stylometry. We cluster the events according to the location of their occurrence, and confirm that the frequency distribution of the events follows power law. We utilize the Word2Vec, a machine learning method for converting words to vectors, to measure the cosine similarity between the events and to understand the relation between the football game events. We construct two networks, one is a transition network that uses the event sequence and the other is a Word2Vec network that learns the data by using Word2Vec. We compare the results to show the significance of the Korean football style revealed in each of the networks.