A Method of Haiku Generation Using Deep Learning for Advertising Generation

We tried various approaches to story generation. One of them was haiku generation using deep learning. The method we tried had several tasks, one of which was to use only word time-series data as training data. Therefore, the result of training can be a model similar to generation based on simple word-transition probability. Consequently, we organized the current tasks, tried to study the features used for training, and generated by applying the learned results. Furthermore, we consider how to use the method proposed in this paper in advertisement generation and in generating different types of stories.