Research On Automatic Generation Of Weibo Short Text Based On User Intention

Due to its convenient,rapid,communication and original characteristics,Weibo has gradually become an important social network platform for people to share short and real-time information.Weibo open platform has created an era of rapid information flow.Now the speed of social development and people's pace of life is accelerating,the speed of information communication flow also accelerated,compared with long complex text content,people prefer to read fragmented information,weibo platform information is more suitable for the needs of The Times,weibo short instinct fully meet the people to accelerate the demand of personal expression and information communication.Weibo short articles can simply express users ' writing purpose.Different users publish different intentions.Ordinary users publish daily insights with the help of platform,enterprise companies with the help of microblog platform,media workers spread news events with the help of microblog platform,and literary and art workers spread knowledge with the help of microblog platform.In order to satisfy the creative intention of different users and assist users to write blog articles,this paper proposes the automatic generation research of microblog short text books based on user intention.The automatic generation of Weibo text belongs to the research of text generation technology.The development of generating confrontation network,cyclic neural network(RNN?LSTM,etc.),encoder and decoder framework,Seq2Seq technology has promoted the progress of text generation tasks,such as automatic abstract,rhythm poetry automatic generation and other related research gradually mature.But the Weibo text belongs to the free text,the text length is different,the language style is diverse,the self-trained model has low robustness,does not apply to the study Weibo text language style.A large-scale natural language pre-training language model(GPT2?BERT?XLnet etc.)was born in recent years,which provides a research direction for natural language generation tasks: pre-training fine-tuning.Therefore,in view of the research of automatic generation of Weibo text,we analyze and summarize the main user intention categories contained in Weibo text,train the user intention recognition and classification model,and use Weibo topic label as a hint.The language model from left to right is used GPT2,and the Weibo text generation task is realized by two-stage fine-tuning method.In the first stage,the pre-training language model is transformed from common language to Weibo text language style,and the second stage fine-tuning language model generates Weibo text based on user intention.The user intention classifier is used to improve the accuracy of user intention prediction generation.Finally,we realize the automatic prediction and generation of Weibo text,and the generated samples meet the social purpose of user topic label and user expectation,and realize the dialogue function of "@ user name " in Weibo text.During the experiment,a small language model GPT2-Chinese suitable for Chinese generation task was selected as the pre-training language model.The experimental results show that the generated sample content is consistent with the Weibo topic and can express the expected social intention of the user to a certain extent.