Research On Keyphrase Generation Of Programming Problem Based On Seq2seq

In the field of subject education,automatically obtaining key information of homework problems,assisting in question comprehension,and achieving automatic answer to questions are hot research topics in the interdisciplinary field of computer and education.Relevant research covers subjects such as Chinese,English,and mathematics.However,in the field of programming education,relevant research is relatively lacking.Automatically obtaining key information of programming problems and achieving automatic understanding of programming problems can help promote the intelligence level of programming education.Therefore,there is an urgent need for a method and technology to automatically obtain keyphrases of programming problems in the field of programming education.Due to the fact that programming problems often use stories or examples to describe program algorithms,a large amount of programming knowledge is directly reflected or contained in the text of the problem.Therefore,this thesis uses keyphrase generation methods to obtain key information that provides guidance for problem-solving in programming problems.This thesis delves into the text of programming problems,constructs a dataset(PPKD)for keyphrase generation tasks in programming problems,and selects the Seq2 Seq model as the basic model.At the same time,a programming problem keyphrase generation method based on Seq2 Seq is proposed to address the characteristics of strong structural composition of programming problems,strong correlation between titles and problem content,strong logical description of problems,and short and small problem texts,as well as the limited data volume and small volume of PPKD.Accurately extracting key information that provides guidance for problem-solving in programming problems can help provide guidance for programming learners on problem-solving strategies,as well as provide basic assistance for intelligent educational applications such as problem classification and intelligent problemsolving.The specific work of this thesis is as follows:1.In order to obtain key information that provides guidance for solving programming problems,a multi mechanism fusion Seq2 Seq model is proposed to generate programming problem keyphrases.Due to the strong descriptive logic of programming problems,the short text of the problems,and the small collective amount of related data,this model selects Bi GRU as the basic network model for the encoder and decoder,and adds attention and copy mechanisms to the decoder of the Seq2 Seq model,quickly capturing useful information in the programming problem text and dynamically expanding the vocabulary to improve model generation performance;In order to meet practical needs,a threshold based beam search filter mechanism is used to adaptively generate keyphrases during prediction output,achieving the generation of an adaptive number of programming problem keyphrases.In this thesis,we constructed a programming keyphrase generation dataset(PPKD),designed hyperparameter experiment,optimization function experiment and ablation experiment.The experiment proves that the generative model of programming problems based on Seq2 Seq used in this thesis performs best,with the P value of the model reaching 0.3856 and the F1 value reaching 0.4915.2.In order to make full use of the structural information of programming problems,a title guided programming problem keyphrase generative model(TG-PPKG)is proposed.Due to the fact that programming problem texts on online platforms typically consist of both titles and main text,which often contain key information that provides guidance for problemsolving,they have the characteristics of strong compositional structure and strong correlation between titles and topic content.Therefore,this model takes both the title and the title+main text as inputs,constructs a sequence pair of[ "title+mian text",keyphrases,title],regards the title as the guiding information for keyphrase generation,and finally quickly obtains more noteworthy information through a one-way attention mechanism from the title to the main text.The experiment proves that the improved model improves the overall F1 value to 0.5209 on PPKD.At the same time,TG-PPKG outperformed other baseline models on two small-scale datasets,Inspec and Krapivin,demonstrating the model’s usability on other small-scale datasets.3.Build a programming practice assistance system to assist students in programming exercises.After training TG-PPKG,this thesis further builds a programming practice assistance system and applies this research content to practical programming teaching.The system functions include question browsing,question search,and question keyword query.At the same time,in order to subjectively evaluate the effectiveness of the model and evaluate the usability of the system,a survey questionnaire was distributed to evaluate the system.The evaluation results show that 63.33% of students believe that the keyphrases generated by the model are accurate,while 31.67% of students believe that they are very accurate;Meanwhile,95% of students believe that the system provides guidance and assistance during programming exercises.