From Quick-draw To Story: A Story Generation System for Kids’ Robot
Lecheng Wang, Shizheng Qin, Menglong Xu, Rui Zhang, Lizhe Qi, Wenqiang Zhang · 2019
From quick-draw to story aims to draw a narrative story based on simple lines children draw. On account of the simple knowledge contained in quick-draw and lack of labelled data, this task is assuredly challenging. In this paper, it is divided into two subtasks. Firstly, a Multitask Transformer Network is proposed to generate the sentence on the grounds of the information of quick-draw. Secondly, we finetune the OpenAI Generative Pre-training Model (OpenAI GPT) to generate a story on the basis of the sentence. However, evaluating the text generation model is difficult. In order to solve with this, a new evaluation metric based on Story Cloze Test is proposed. Our evaluations indicate the prominent results our method has achieved in this task.