Research On Text Deep Analysis Based Storyline Generation

With the development of computer technology,Natural Language Generation(NLG)has gotten more and more people's attention.Natural language generation can be used to generate natural language sentences that are similar to human expressions.Text paraphrasing is a branch of natural language generation technology.It is used to express the meaning of the original sentence in another way and applied to many fields such as machine translation and question answer system.Although paraphrasing has already developed for a long period of time.There is relatively little research on paraphrasing generation by movie dialogue information at persent.Describing the stories of the movie in an objective manner and generating the storyline information of the dialogue can make users effectively understand the movie dialogue information.However,due to the colloquial and subjective features of dialogue information,it is difficult to generate paraphrasing using the dialogue.This thesis proposes a story generation method based on text deep analysis,which can be divided into natural language conversation extraction and movie storyline generation.For the dialogue information in the movie subtitles,this thesis designs a natural language dialogue extraction method.Through the disfluency detection and processing for movie dialogues,the redundant information in the sentences is removed;through the anaphora resolution for the dialogue,the intelligibility of the sentences is enhanced;through the combination of the question sentences and the answer sentences,it can form a close relationship between the questions and sentences,also prevents the separation of complete information.In order to organize the different stories in the movie,this thesis adopts the semantic relationship to divide the movie dialogue,and adopts the semantic-based clustering method to generate the dialogue scene.On the purpose of reserving the semantic information of the sentences in the dialogue and abandoning the expression of the original sentences,this thesis extracts the core elements of the sentences in the dialogue.For obtaining sentences with core elements,this thesis designs a method for acquiring the sentences including core elements of dialogues.This thesis uses multiple search engines to obtain related sentences,selects these sentences based on search locations and semantics,and finally obtains related sentences with core information.For the sake of simplifing related sentences and generating the descriptive information of the movie,this thesis designs a paraphrasing generation model based on the Attention mechanism.By using the Attention mechanism,the non-important words and phrases can be filtered out,making the model can pay more attention to important words and phrases.Also,this thesis designs the training and generation method of the movie storyline generation,which can be used to generate the final movie dialogue storyline.Finally,some experiments are conducted on the method proposed by this thesis through data sets.First,this thesis identifies the basic six movie categories.And 10 movies for each 6 categories from the movies with high ratings on IMDB are selected.These movie subtitles are used as dialogue information,and the synopsis information in IMDB is used as the standard paraphrasing data set.Then,this thesis presents an example of the experimental process of natural language conversation extraction and movie storyline generation.By using the ROUGE-1,ROUGE-2 and semantic similarity assessment methods to evaluate the experiment,it can be found that the method proposed by this thesis has higher performance than the LEAD,MMR and Text Rank methods.