Research On Automatic Generation Of Soccer News Based On Textual Live Broadcast

With the development of the sports live broadcast platform,a large number of textual live broadcast have appeared in the live room of soccer matches.At the same time,the rise of mobile terminal reading has greatly increased the demand for soccer match news.Textual live broadcast is a colloquial description of the real-time process of the competition,which is lengthy and fuzzy.The soccer match news is a general report on facts in soccer matches,generally short and focused on key events.It will take much time to write the soccer match news and affect the real-time performance of the news.Generally speaking,the textual soccer match live broadcast contains most of the information in a match.Therefore,the focus of this thesis is to generate news of soccer matches based on important information in textual live broadcast.This thesis regard the soccer news generation problem as a text summarization task.However,the traditional method of text summarization will have problems such as loss of important information and low readability in soccer field.To solve these problems,this thesis divides the news generation of soccer matches into three parts: Generate the summary and ending of the soccer match news based on player data,team data and sentence template;Use classification models to extract sentences from annotated textual live broadcast;Paraphrase generation based on extracted sentences and sequence to sequence models.This thesis proposes a soccer match news generating method based on textual live broadcast and attention encoder-decoder model.Firstly,this method extracts sentences from the textual live broadcast based on the text features and classification models such as extreme gradient boosting,convolution neural network.Secondly,using the encoder-decoder model based on attention model and bi-directional long-short term memory network to generate paraphrase for extracted sentences.Finally,concatenate the generated sentences,improve the readability of the news in the premise of recalling important events.This thesis extracts and selects textual features suitable for soccer field.Combining multiple machine learning models with news generating tasks of soccer matches.Using rule based sentence templates to enhance the quality of soccer match news generated.The experimental results show that it is feasible to use the method proposed in this thesis to generate soccer match news from textual live broadcast.