Learning to Rank Model Performance and Review with Pairwise Transformations
Ajinkya Vijay Koshti · 2021
Learning to rank is an application of machine learning that is finding increasing use in information retrieval systems. Various applications such as document retrieval, webpage ranking, sentiment analysis, and online advertising use one or the other kind of learning algorithm to return a list of instances ranked in order of the quality of fit. Learning to rank also finds application in sports analytic; especially in activities like football scouting, where club representatives attend football games worldwide to collect intelligence on potential recruits. In this project, learning to rank method is applied to automate the process of football scouting by analyzing the player database to generate a list of desired number of footballers that could be recruited based on given criteria. This becomes an important application as player recruiting is a competitive business and having a list of options ranked on given criteria can help in making faster and better decisions. Learning to rank technique is used with pairwise-approach that makes it possible to use state-of-the-art supervised algorithms for generating predictions. Borda count technique is applied to convert the pairwise predictions into a ranked list.