T4S: Two-Stage Screenplay Synopsis Summary Generation with Turning Points.

Depei Wang,Wenyi Sun,Cheng Luo,Dachang Liu,Ruibin Mao,Ruifeng Xu

DOI: https://doi.org/10.1007/978-3-031-51671-9_6

2024-01-01

Abstract:The recent advancements in neural network models and the availability of vast amounts of data, automatic summarization technology has become one of the primary solutions for dealing with information overload and pinpointing key information. Unlike narrative text, movie scripts consist of sequences of scene descriptions, and directly compressing the text may lead to truncation of plot-relevant content. Furthermore, movie script summarization tasks lack datasets that align script content with plot summaries. Therefore, this paper proposes a two-stage method T4S for generating plot summaries of movie scripts. First, the GraphTP model is employed as an extractor to extract key turning scenes from the scene text sequences. Second, an unsupervised text matching method is used to obtain text pairs that match scenes with plot summaries. Finally, a generator, utilizing an efficiently fine-tuned LLM, rewrites key scene text and concatenates it to form the final plot summary of the movie script. The results of the implementation show that the proposed method in this paper outperforms baseline methods.