Research On Graph-based Method Of Trending Event Storyline Generation

With the rapid development of Internet technology,social media platforms,such as Weibo and Twitter,have become more and more effective sources of information for people to learn about recent trending events and their views.Weibo or other short essay delivery platforms can provide users with real-time and effective information that traditional media cannot provide,the research on Social Media Mining are becoming more and more popular.For trending events with tortuous plots and high attention,it is very important to quickly obtain information about their development process.In this paper,the storyline generation of trending news events is to generation the development context of social events or entertainment events in reality.Specifically,the model is used to extract the effective information of development of events from the chaotic short text data and organize it into a concise and accurate sequence of stages of the event.However storyline generation is very challenging due to the short length of posts,lack of context,excessive redundancy of information,and diversity of vocabulary.The existing models can be mainly divided into two categories,Clustering-based and Graph-based optimization approaches.Although these models can achieve effective information extraction to a certain extent,there are still problems: the accuracy,integrity and coherence of the storyline cannot be guaranteed,and there is no evaluation method specially used for storyline of events,so the quality of storyline cannot be comprehensively evaluated.This paper proposes the following two models for the above problems and proposes a new evaluation index for synthetical choroid integrity and coherence.(1)Event storyline generation model based on HWB stage detection algorithm is based on the graph optimization algorithm.On the basis of ensuring coherence,combining with the difference in the distribution of hot words in Weibo data between stages,the HWB stage detection algorithm is proposed to ensure the integrity of context.Moreover,the community discovery algorithm is used to further filter the data of the input of the graph optimization algorithm,so as to keep the data diversity and reduce the data volume.An improved graph optimization algorithm is proposed to optimize the calculation method of the weight of the points in the graph so as to keep the diversity of data.Thus,an integrity,coherent and accurate sequence of events can be generated.We designed a comparison experiment on the real data set to prove the validity of each module of the model.At the same time,three existing models were selected for comparison to prove the validity of the model as a whole.(2)Event storyline generation model based on Dynamic Keyword Graph stage detection algorithm is proposed to further improve the integrity of the storyline.HWB based stage detection algorithm relies on the influence of event stage.To solve this problem,we propose an algorithm that applies graph structure to stage detection,named Dynamic Keyword Graph stage detection algorithm.This algorithm makes use of the different construction of the keywords between different stages,and do not rely on the influence event development stage.The model further improves the integrity of the context by establishing dynamic keyword graph for phase detection.In addition to Dynamic Keyword Graph stage detection algorithm,a graph optimization algorithm based on sentence level is proposed to filter redundant sentences to improve the accuracy and readability of the stage.Similarly,we design comparison experiments prove the validity of the model as a whole and each module of the model.(3)SLEU,based on the idea of BLEU,is designed based on sentence level to evaluate the quality of the storyline,which comprehensively considers the integrity and coherence,instead of directly regarding the storyline of the event as a text summary as in the existing work.The results of SLEU's evaluation are more consistent with human evaluation habits and user survey results.