The American publisher recently filed a patent centered on an obvious need for anyone who has ever tried to play competitive games online, namely a matchmaking pertinent. Using the method of deep learning – that is to say the use of algorithms simulating the functioning of a brain via an entire artificial neural network – the technology in question will be based on the analysis of the behavior of players. If the company has tried different approaches in previous years, in particular via the skills calculation method Rating often used in chess, this new approach could compensate for the weaknesses of these not necessarily optimal systems.
Without going into too complex details, the principle here is based on the fine analysis of the players with regard to their overall statistics in the course of the game or the games in which they have participated. This then makes it possible to sort according to the profiles with the best individual capacities, adapted to a particular context; in order to compose balanced and efficient teams. Each player is therefore assigned a sort of score, sum of raw figures – for example the ratio of dead / killed in an FPS – and behavior in match (regularity, reaction to a change of role, etc.). What answer, for the moment in theory, to this recurring concern of the matchmaking as a source of disengagement.