Research And Implementation Of Key Technology For Intelligent Generation Of Conference Summary

In this era of big data,with the explosive growth of Internet information,information overload has become an inevitable problem,At the same time,with the development of artificial intelligence technology and the gradual increase in egovernment and online conferences,it is important to extract and condense central content of the conference to form the conference summary,which is of great significance for improving the automation level and work efficiency in the office field.The WEAI system is a typical application in the office field.It realizes office automation with the support of artificial intelligence technology and improves the intelligence level of the office system.As a summary of the content of the conference,conference summary is a document that is refined on the basis of the meeting records that can completely and truly reflect the theme and key information of the meeting.This paper studies the extractive summarization technology for conference summary in the WEAI system,and has the following research contents:1)Extractive summarization algorithm based on improved graph sorting.To study the role of different word vectors in the semantic representation of text,introduce the BERT pre-training model and compare it with the word2vec model.The K-means text clustering algorithm is used to divide the text to obtain multiple subject content in the text,and optimize it from both the determination of the initial centroid and the number of adaptive classifications.Finally,consider the contextual structure characteristics of sentences and texts,filter sentences by sentence length and number of clauses,and calculate the weights of conference name similarity,sentence position,keywords,etc.to adjust the weight of vertices in the graph model.Complete the improvement of the TextRank algorithm and verify its effect through experiments.2)Sentence extraction model based on fine-tuning BERT and attention mechanism.On the basis of fine-tuning the sentence vector of BERT,the document vector is extracted by the long and short-term memory network,and the sentence vector and the document vector are used to jointly represent the text context semantics,and the speaker identity is introduced as external information.Sentence extraction is based on the sequence-to-sequence model,and attention mechanism is added to select key sentences more accurately,and the influence of multiple decoder on the model is compared.3)Design and implement WEAI system based on the proposed algorithm,which can process conference text records and generate summary,introduce the overall system design and various functional modules,and display system functions.