Indie Games Popularity Prediction by Considering Multimodal Features

Yu-Heng Huang, Wei-Ta Chu

Abstract: We present a popularity prediction system for independent computer games (indie games), by jointly considering visual, text, and metadata information. An indie game dataset is first collected and labeled. According to the number of sales, we label an indie game as popular or not. Different types of information is extracted by specific feature extractors, and then is fused to construct a neural network-based classifier. We demonstrate that jointly considering multimodal information yields promising performance. In addition, we show that, with helps of state-of-the-art feature embeddings, the proposed method outperforms the only existing SVM-based method.