Predicting game popularity through graphic modeling

The gaming industry has been a profitable area. But recently given the saturation of the market where there are more than 100,000 games on steam, each game must under go careful planning prior to its release. The planning often involves making sales predictions given various features of the game. Our model incorporates a graphical neural network in making binary classification using data provided by the Steampower APl. There exists prior methods such as deep factorization machine, random forest, and graph convolutional network in targeting similar issues but none of them have been used to analyze game popularity. Referring to Tab GNN, we first implemented a multiplex graph embedding using the features obtained from Steam power then applied GCN conv to enhance the learning of graphical embedding. These embeddings were then fed into a 3-layer-DNN to make binary predictions. Experiments with other baseline models were conducted showing that our model is possible in making relatively more accurate results in predicting game popularity.