Groupe de travail

Discovering tactics from team sports data

par Albrecht Zimmermann

Europe/Paris
S3 351 (Sciences 3)

S3 351

Sciences 3

Description

Team sports tactics are about who (which player and/or player role) does what (in terms of legal actions, which includes moving around) where on the pitch, court or whatever the playing field is called, potentially modified by when (in the shot clock, game clock etc).I used my presentation to quickly touch on those different aspects (excluding time) by summarizing a number of works from the literature. The lines are not always clearly drawn: some papers mix what athletes do with where it happens or with who does it. 

The "what" part is arguably the most complex one and the papers I touched on all use what I've called the "vocabulary" of possible actions/movements to describe what teams and individual athletes do. Some do so rather explicitly, using topic models originally developed for text document analysis, others learn or predefine patterns that occur in certain situations or for certain teams. Some learn the vocabulary from the available data, others predefine it and "only" learn the "phrases" that are being formed.

When it comes to the "who", finally, network-based modeling proves to be very powerful and allows to explore "what if" scenarios that would occur if one put different line-ups in the field.