Research On Automatic Generation Of Sports News Based On Neural Network

The live scripts for sporting events use the live data in sports competition as information sources,and use the network platform as the medium to broadcast the sports of live to sports fans in text form.As the way of live broadcast of texts can provide another way to watch the game dynamically for Internet users who cannot watch sports games through traditional media,it has gradually become a useful complement to the live video-based sports competitions,and has received a lot of attention and participation from Internet users.The live script describes the progress of a game through text.As the game continues,the live script updates the report in real time according to the match between the two sides to enhance the user experience.At the same time,it also provides first-hand information for sports reporters to help them write high-quality sports news based on this.The large-scale live scripts for sporting events reflect the real-time process of the game quickly and timely,but it is often time consuming and laborious for a journalist to write sports news.In view of this,we propose a neural network model which automatically generates the sports news corresponding to the sports live scripts.The model avoids the limitation that the traditional model is too dependent on the hand-crafted features.Besides,it can also consider the script sentence-level local information and global information and semantic relevance between script sentences and corresponding news content in the scripts,which realize the formation of sports news under joint modeling.Based on the sports news content generated by the neural network extraction method,the end of the news is further optimized.The end of sports news generation is improved based on the model in Abigail See's paper combined with the task characteristics of the sports news field.We mainly extract keywords from the end of the live script to improve the attention mechanism in the original text,and realize the sports news combined with the extractive and generative methods.The experimental results on the open data set verify the feasibility and effectiveness of the proposed method.In addition,we also try to generate the title of sports news based on rules and templates to extract the key content of it.