Research And Implementation Of OJ Question Analysis System Based On Deep Learning

With the increasing emphasis on programming education by the country and society,people pay more and more attention to the cultivation of students’ programming ability in information technology courses of primary and secondary schools.More and more students begin to use the Online Judge Platform to practice programming and algorithms.Compared with college students and adults,students need more guidance when using OJ Platform.However,some guidance designed by popular OJ Platform require a lot of manpower and material resources for data labeling,which is time-consuming and labor-intensive,and the guidance functions are still not perfect.Students still have trouble in using OJ Platform.Adding more intelligent and convenient functions to OJ platform will help improve students’ enthusiasm for learning.In this paper,we conduct research on the intelligent analysis of the massive question bank on the OJ platform,which designs two convenient functions:OJ question abstract generation and auto-generating OJ test paper.Functions can not only guide students to better understand the meaning of questions,but also provide students with a way to simulate competition.In the research of the OJ question abstract model,an adaptive sentence segmentation algorithm based on symbol priority is proposed,and the dataset is constructed by combining automatic abstract and manual screening.Based on the baseline model BertSum,according to the characteristics of OJ topics,we improve an extractive abstract model CBertT,and design a word vector convolution processing module and a vector fusion module based on TextCnn.Finally,the effectiveness of the improvement is verified by comparative experiments.In order to improve the quality of auto-generating paper,the genetic algorithm was used,and the dataset was constructed by web crawler.Firstly,a segmented real number coding scheme with redundancy is proposed to encode the chromosome of OJ questions.Then,we propose a similaritybased calculation method for the fitness of OJ questions,and designed the selection,crossover and mutation genetic operator modules in detail.Finally,relevant experiments were carried out to analyze the quality of auto-generated papers.Finally,using software engineering theory,we design and implement an OJ question intelligent analysis system based on the microservice architecture,which provides OJ question abstracts and auto-generating OJ test paper.In addition,we conduct functional tests,performance tests and robustness tests to verify that system has achieved the expected effect,and the system can help students to use OJ platform more conveniently and quickly.