Continuing their research on the data set of Tomb Raider: Underworld gameplay metrics, the researchers use Deep Learning to discover four different types of players.
Anders Drachen, Alessandro Canossa, Georgios N. Yannakakis
We present a study focused on constructing models of players for the major commercial title Tomb Raider: Underworld (TRU). Emergent self-organizing maps are trained on high-level playing behavior data obtained from 1365 players that completed the TRU game. The unsupervised learning approach utilized reveals four types of players which are analyzed within the context of the game. The proposed approach automates, in part, the traditional user and playtesting procedures followed in the game industry since it can inform game developers, in detail, if the players play the game as intended by the game design. Subsequently, player models can assist the tailoring of game mechanics in real-time for the needs of the player type identified.
In this paper, I was introduced to the Deep Learning algorithm: Self Organizing Maps (SOM).
After doing unsupervised learning with SOM, the researchers found that players are grouped into 4 groups:
Veterans: Players that die very few times; their death is caused mainly by the environment and they complete TRU very fast. These players’ hints requests vary from low to average.
Solvers: Players that die quite often mainly due to falling; it takes them quite a long time to complete the game, and they do not appear to ask for puzzle hints or answers.
Pacifists: Players that die primarily from active opponents. The total number of their deaths varies a lot but their completion times are below average and their help requests are minimal indicating a certain amount of skill at playing the game.
Runners: Players that die quite often and mainly by opponents and the environment. These players are very fast in completing the game (similar to the Veterans) while having a varying number of help requests.
Its quite interesting that a game like TRU, that is linear in its story and challenges, ends up having four different playstyles. I happen to see myself mostly as a solver, however, I do act sometimes as a Pacifist. This makes me wonder if playstyle actually evolves or change as I play the game.
Lastly, I wonder if we can figure out the player’s playstyle as he plays the game, and then adjust the game to be more of his liking. For example, if he is a Pacifist, we reduce the difficulty of the game-controlled opponents.
The paper can be accessed at: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.456.4596&rep=rep1&type=pdf