Automatic Text Generation System For English Scientific Papers

With the increasing investment in scientific research in modern society,more and more papers are produced every year,but English paper writing is very difficult for non-native English speakers.Writers not only need to clearly express the experimental process and results,but also use English vocabulary correctly.In recent years,natural language generation technology has developed rapidly,and great progress has been made in both theory and practical application.this paper studies the application of mature natural language generation technology in English paper writing,and is dedicated to improving writing efficiency.This paper first analyzes the problems in the implementation of the system,and introduces the model algorithms to solve the problems.Aiming at the imbalance between the encoding and decoding sides caused by the use of pretrained language models in paper title generation,this paper adopts the T5 model suitable for generation tasks and uses paper data for fine-tuning;in addition,for the problem of paper abstract generation,This paper improves the graph2 text model and designs a complete and humanized operation process based on the information extraction tool Dyg IE and graph database based on the knowledge graph structure used by the model.the system uses the Word2 Vec word vector model,Stanfor Core NLP part-of-speech analysis algorithm and The word similarity matching algorithm realizes the word polish function of word granularity,and uses the LASERTAGGER sequence tag model to realize the polish function of sentence granularity.Subsequently,the system is divided into4 core functional modules,which are developed using Spring Boot and micro-service structure,and the design process of each module is introduced.The system is mainly divided into title generation,abstract generation,text retelling and background management modules,and is equipped with a corresponding database as the underlying data support.The system designs a series of text generation models for paper writing based on scientific papers,which can help users generate paper titles,abstracts and polish sentences quickly and conveniently,reduce the burden of users writing.The system builds front-end interface client in the form of Word plug-in,which fits actual writing scene and provides humanized interaction method,allowing concentrate on writing and improve work efficiency.