Abstract
This work has been inspired by a recruiting system that a Spanish football club created with artificial intelligence making use of both numerical and textual descriptions of players. However, the news piece presenting this approach did not give many details. Within this context, I was curious to know whether ChatGPT could provide me with some more information, which further led me to the idea of trying to develop my own player matching procedure. Following several constructive exchanges with ChatGPT and after performing some successful tests, I present the algorithm in this paper. The main questions that this research aims to discuss are as follows: how ChatGPT can be utilized in autonomous research projects, particularly in sports; how to overcome the lack of public football data, also considering textual information about players; and how to support the expert knowledge of recruiters and coaches with data-based tools. Indeed, one limitation when carrying out football analysis is the difficulty accessing public data, not to mention unstructured texts. I propose to use data from video games to mimic recruitment tasks. The methodology behind the algorithm is based on text embeddings and the Gaussian kernel to work with the text and numbers that represent each player, respectively. A new R package called fawir that includes the recruiting algorithm and a data set to reproduce the results is also presented.
Get full access to this article
View all access options for this article.