Optimizing Multi-feature Dependent Story Generation Model

Story generation is an interesting but challenging problem in natural language processing. Although large-scale language models have succeeded in generating fluent texts, they still have problems of repetition, inconsistent logic and lack of long-term consistency. This is because the input information of the task is limited, which makes it difficult to select the content from the limited information to generate the model. In addition, the knowledge-driven methods lack consideration of specific story background, and the introduced knowledge will have a negative impact on the generation. Therefore, we design a story generation model (OMFD) that optimizes multi-feature dependency. The model adopts a dual-encoder structure, and realizes the separate processing of text and knowledge information in the coding stage. The decoder adds a knowledge management module, which is responsible for selecting reasonable knowledge to guide the generation. The experimental results show that our method is superior to various baseline models, indicating that our model can generate stories with richer vocabulary, smoother sentences and more coherent content.