Deep learning contextual models for prediction of sport event outcome from sportsman’s interviews

Proceedings of Recent Advances in Natural Language Processing, pages 1240–1246, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_142 1240 Deep Learning Contextual Models for Prediction of Sport Events Outcome from Sportsmen Interviews Boris Velichkov and Ivan Koychev Faculty of Mathematics and Informatics, Sofia University ”St. Kliment Ohridski”, Sofia, Bulgaria bobby.velichkov@gmail.com koychev@fmi.uni-sofia.bg Svetla Boytcheva Institute of Information and Communication Technologies, Bulgarian Academy of Sciences, Sofia, Bulgaria svetla.boytcheva@gmail.com Abstract This paper presents an approach for pre- diction of results for sport events. Usu- ally the sport forecasting approaches are based on structured data. We test the hypothesis that the sports results can be predicted by using natural language pro- cessing and machine learning techniques applied over interviews with the players shortly before the sport events. The pro- posed method uses deep learning contex- tual models, applied over unstructured tex- tual documents. Several experiments were performed for interviews with players in individual sports like boxing, martial arts, and tennis. The results from the conducted experiment confirmed our initial assump- tion that an interview from a sportsman be- fore a match contains information that can be used for prediction the outcome from it. Furthermore, the results provide strong ev- idence in support of our research hypoth- esis, that is, we can predict the outcome from a sport match analyzing an interview, given before it. 1 Introduction 1.1 Motivation The problem of predicting sports results is very challenging and is widely explored in artificial intelligence (AI) (McCabe and Trevathan, 2008). This task requires application of complex algo- rithms over a huge variety of heterogeneous types of features. Classical decisions are based on sta- tistical and probability models. The AI techniques used to solve this task are based on machine learn- ing (Keshtkar Langaroudi and Yamaghani, 2019) and data mining (Haghighat et al., 2013). This is due to the lack of large datasets with previous re- sults for players and games. Sport is a very dy- namic area and players are active for a relatively short period of time. There are also limitations to the predictability of sports outcomes data over a long period of time. Team sports are more difficult to predict be- cause different team members are selected to play games and even during the game several changes are made to the team, several penalties and injuries to the players that can have a huge impact on the end result. Such predictions in sports rely on many features and time models over a long period of time. In this way, we are able to tackle the task of predicting sports results only in individual sports, such as tennis, boxing, mixed martial arts (MMA) and etc. 1.2 Related Works and Methods The task of prediction of the sport results can be solved as a classification problem. Na¨ıve Bayes (NB) (McCallum et al., 1998) is the simplest method of classification. It can show good results even for small sets of training data, as is the case with the sports prediction task. The main draw- back of the approach …

Content truncated to 500 words