In the film industry, script creation and narrative quality are crucial to the success of a film. Based on reinforcement learning, Script2Graph method is used in this paper to solve the problem of highly correlated information extraction and management framework in the plot analysis of large corpus. The method first converts unstructured text into structured information by parsing the script's unique format. Secondly, the NLP information extraction pipeline is constructed to obtain text world elements. Finally, we introduce the attribute information of the edge into the text world system, construct the triplet of script knowledge, and perfect the framework of information organization. Experimental results show that compared with the baseline method, Script2Graph based on the pipeline method can improve the accuracy of script analysis, the coverage rate of event trigger words and the accuracy of emotion classification by 1.85%, 4 times and 7.3%, respectively. This method can automatically obtain the diversified information in the movie script, has a strong information integration ability, and can support the downstream task well. Therefore, the introduction of reinforcement learning can not only use reinforcement learning algorithms to evaluate and optimize different plot trends, so as to achieve personalized script generation. In addition, reinforcement learning can also help creators explore new narrative structures and character development ways to promote innovation and progress in film storytelling.