Deep learning for video game genre classification

In this paper, we propose a new multi-modal deep learning framework with visual modality and textual for video game genre classification. The proposed consists of three parts: two deeep networks data imaginary data, feature concatenation algorithm, then softmax classifier. Video covers descriptions are usually the very first impression to its consumers they often convey important information about games. classification based on cover description would be utterly beneficial many modern identification, collocation, retrieval systems. At same time, it is also an extremely challenging task due following reasons: First, there exists wide variety genres, which not concretely defined. Second, vary in different ways such as colors, styles, information, etc, even games genre. Third, designs may external factors country, culture, target reader populations, etc. With growing competitiveness industry, designers typographers push limit hope attracting sales. computer-based automatic systems become particularly exciting research topic recent years. contribution paper four-fold. compiles large dataset consisting 50,000 from 21 genres made images, text, title text information. image-based text-based, state-of-the-art models evaluated thoroughly developed efficient scalable both images texts. Fourth, thorough analysis experimental results given future works improve performance suggested. show that outperforms current or text-based models. Several challenges outlined task. More efforts resources needed order reach satisfactory level.