Research And Implementation Of Submission Recommendation System Based On Improved Contrastive Learning

In recent years,the research environment has continued to improve,academic conferences are held more and more frequently.Academic communication and creative efficiency have increased substantially,and a large number of academic groups have emerged.The number of applicants for master’s and doctoral degrees has also reached an unprecedented number,and the group of published papers has grown.Article writing and submission has become more common,with submission being an essential part of publishing a paper.The need for efficient selection of appropriate journals for submission has grown stronger.Especially for those working in the field of research,it takes a lot of time to select the right journal for submission of research results.Unsuitable submissions consume the time and energy of scholars as well as the time and energy of editors and reviewers.The existing submission recommendation system ignores important contents such as research methods and academic opinions contained in cited literature.The recommended journals are not comprehensive enough,and the recommendation effect is poor.In this thesis,we address the above problems and conduct a research on the submission recommendation system based on improved contrastive learning,which includes the following two aspects of research:(1)A submission recommendation algorithm based on improved contrast learning was proposed.Contrast learning prefers journal content of the same or similar length to be semantically more similar,resulting in low accuracy of the algorithm.This thesis uses random repetition and random deletion strategies to solve this problem by improving the contrast learning algorithm and combining it with a network of citation relations for submission recommendation.Supervised learning is used to close the distance between a paper and its matching journal in the embedding space.The paper information and the journal information data belonging to them are processed into sample pairs.Using a pre-trained BERT language model as an encoder,the obtained data is fed into the contrast learning framework to fine-tune the model parameters.The semantic distance of each journal is expanded using unsupervised contrastive learning approach.It makes standard dropout.The same input text is randomly repeated and randomly deleted.Then input to the encoder as expanded data,with the information of the journal itself as the label and the other journal information in the input text as negative samples.The submission recommendation algorithm based on improved comparison learning distances each journal in the embedding space and improves the submission recommendation effect.In-depth research on the submission recommendation algorithm is carried out by learning the feature links between journals and papers in massive data and incorporating citation relationship networks.The effectiveness of the submission recommendation algorithm based on improved contrastive learning is verified through comparison experiments with other mainstream algorithms.The accuracy rate of the proposed algorithm is improved by comparison.(2)A submission recommendation system was developed.In this thesis,a requirement analysis is conducted to determine the core function of pushing journals with high matching degree by analyzing the content of user’s papers to be published.The system is analyzed for functional and non-functional requirements.Each functional logic and subsequent extension,maintenance,and security are introduced in detail.The overall architecture is constructed based on the analysis.The application system is built using the B/S structure with the MVC model.The database design part strictly follows the paradigm rules.The system designs the database according to the application requirements,while fully considering the possible future expansion and changes.Finally,the software and hardware development environment for system operation is configured.Each functional module page is displayed,and the performance of the system is tested for each functional page and after operation.Through a variety of verification methods to detect possible vulnerabilities of the system in a timely manner,to ensure the stability and reliability of the system.The test results show that the system achieves the expected goal.It can give full play to the advantages of artificial intelligence.It pushes matching academic journals for users efficiently,optimize the submission process.It improves the hit rate of submission,shorten the submission cycle.It brings practical application value to users.