Automatic Creation of LSTM Game Level Based on Attention Mechanism

Procedural Content Generation (PCG) is a game design technology, which uses automatic generation technology instead of traditional manual creation, and this method is also applicable in level generation of platform games. However, for games with a single mode, such as platform games, it will make players feel boring to pass the same level repeatedly. How to improve the interest of games is one of the problems faced by platform games. In this context, this paper proposes a LSTM-based game level automatic creation method to generate a variety of game levels. In order to facilitate the generation of platform games, the game level is transformed from image file to text file by establishing mapping table between maps and symbols, and the input data is preprocessed to some extent. LSTM model can be used to predict the generation of game levels, and the addition of attention mechanism improves the accuracy and coherence of game level generation. This method effectively solves the problem that simple platform games can't attract players' attention for a long time.