Research And Implementation Of Question Recommendation For OJ Platform

With the popularization of information technology education,various online judge platforms have been widely used in primary and secondary school teaching.However,the questions on traditional platforms are not classified by algorithm topics.For autonomous learning,students hava to randomly select questions from the list,which is inefficient.In order to improve users’ learning experience and efficiency,we do some research on question classification and recommendation for OJ platform,aiming to realize automatic classification and provide personalized recommendation services.OJ question information mainly includes description text and solving program.In order to extract the semantic and structural information of the solving program,we design three program representation methods based on syntax tree path to realize the feature embedding of solving program,and then adopt the multi-channel feature fusion method based on self attention mechanism to obtain the vector representation of source program.In the research of OJ question classification,we design the model based on TextCNN.According to the description and solution program of the topic,the qustions are classified by algorithm topics.Comparative experiments show that the accuracy of code2vec is improved by 6.1%compared with the original model,which verifies the effectiveness of the program representation method based on syntax tree and feature fusion.Based on the classification for questions,we propose a recommendation model integrating collaborative filtering and user behavior sequence.For heterogeneous information of users,a multimodal global attention mechanism is designed to extract high-dimensional cross features.And for the sequence information of user behavior,we design a sequence feature crossover module based on AFM network.Finally,the effectiveness of multimodal global attention and sequence feature crossover is verified by public data sets-Moivelens and Amazon Book,respectively.The online AB test and offline comparison show thar our model for OJ question personalized recommendation is effective.Finally,based on self-built OJ platform-CodeOJ,a personalized recommendation system for OJ questions is designed and implemented.The stability of the system is verified through function and performance tests.In high concurrency application scenarios,automatic classification of questions and recommendation of similar questions are relized.