A Deep Learning Supported Sequential Recommendation Mechanism for Ban-Pick in MOBA Games

Abstract

Existing research on MOBA (Multiplayer Online Battle Arena) games focuses mainly on the composition of a lineup being formed during ban-pick, a process designed for selecting champions in a fixed temporal order before the game starts. Although it is apparent that the order by which the lineup is formed may have a crucial impact on the results of the entire game, current research has given little attention to the temporal order thanks to insufficient labeled data required for such exploration. We planned to investigate the potential consequence of taking account of such an order during lineup formation. To this end, we collected data from open-source platforms and attached temporal information to these data. Moreover, we designed a sequential recommendation mechanism of champions for ban-pick. Our recommender was enhanced by drawing on NLP techniques, Bi-LSTM models and knowledge graph. It is shown by our experimental results that the proposed recommendation mechanism is effective in terms of both the coverage rate of champions and user satisfaction.