Design And Implementation Of Paper Classification System Based On Graph Data Few-shot Learning

Scientific research directions are becoming increasingly rich,and the accurate classification of papers can bring great convenience to scholars doing research.The current thesis classification and retrieval system suffers from the problem of large classification granularity and inability to retrieve accurately.When the system deals with papers in areas of cold research,the data volume of such papers is small,and ordinary classification models cannot achieve satisfactory results,which not only reduces the work efficiency,but also brings bad user experience to researchers.This project designs and implements a classification system for small sample papers based on graph data,which solves the problem of inaccurate classification when dealing with papers with small sample size.The main work accomplished in this project is as follows.First,a scheme for classifying small-sample papers based on graph data is designed.The scheme first uses the features of each paper abstract as nodes and the citation relationship between papers as edges to build a citation network,then uses graph neural network as a training tool,a smallsample algorithm based on meta-learning as the training process,and an nway k-shot multi-task structure as the training method,and finally trains different classification models according to different application scenarios,and applies the models to the system.For the problem of small sample size,the solution applies the few-shot meta-learning Reptile algorithm,which speeds up the training process of the model,avoids the secondary update of the gradient on the MAML algorithm,and improves the training efficiency,while achieving the same training effect.Secondly,a small sample paper classification system was designed and implemented.Users upload papers according to their needs,and the system selects different classification models to classify them according to the characteristics of the uploads,and finally gives the classification results back to the users.The front-end of the system is based on the Vue.js framework and the back-end is based on the SpringBoot framework to achieve front and back-end separation.The back-end distributed services use Dubbo RPC framework to call each other.The database uses MySQL and Redis for storing different types of data,and a file server for storing papers uploaded by users.The paper describes the requirements analysis of the small sample thesis classification system,proposes the database design and system architecture design,and details and implements the thesis upload module,the thesis classification module,the thesis information retrieval module,and the system management module.The final results of the system testing show that the system meets the expected results and has value for use.