Research On Personalized Question Recommendation Based On Knowledge Graph

With the development of Internet technology and online education technology,the scale of online questions has grown rapidly.At present,many questions websites present the characteristics of a large number of questions but a single screening structure,which leads to problems such as choice lost and poor learning pertinence.Most of the existing question recommendation methods only use the investigated knowledge points to recommend questions to students,ignoring the related but not investigated knowledge points,which can not find the students' knowledge loopholes and weak knowledge points and affect the accuracy of question recommendation.Therefore,based on the knowledge graph of junior high school mathematics curriculum,this thesis designs a personalized question recommendation method according to the students' mastery of knowledge points in the learning process.The specific research work is as follows:Firstly,taking junior high school mathematics curriculum as the research object,according to the characteristics of clear hierarchy and clear relationship,this thesis adopts the top-down approach,combining with manual and automatic means to build the curriculum knowledge graph.Based on the electronic textbook of junior high school mathematics PEP and the latest junior high school mathematics curriculum standards,Protege software is used to construct the pattern diagram and knowledge ontology.According to the constraints of the model layer,manually extract the data instances of entities and relationships,and match them with the model layer,thereby generating a higher-quality knowledge graph library of chapter-section-knowledge points three-tier structure.Manual construction can ensure the quality of the knowledge graph,but it consumes a lot of manpower and time,and the scale of construction is relatively small.In order to expand and iterate the existing graph,the crawler technology is used to obtain the original corpus data of Baidu Baike,and through automatic processes such as data preprocessing,entity extraction and relation extraction,the verified triples are merged with the manually constructed knowledge graph to obtain the final knowledge graph.Secondly,based on the curriculum knowledge graph,the learner is modeled,and a personalized question recommendation method based on cognitive diagnosis and intimacy between knowledge points is designed.On the other hand,a personalized knowledge subgraph is generated for individual students based on the learning goal and knowledge graph.The entities and relationships are represented by vectors through the TransR,and the intimacy between knowledge points is calculated by combining the entity semantic vectors and relationships.According to the acquired cognition level and the intimacy between knowledge points,the recommendation rules are formulated to recommend suitable questions to students.The experimental results show that the designed question recommendation method model enhances the interpretability and accuracy of the question recommendation,and makes the recommended results traceable.At the same time,it enhances the pertinence of recommendations,makes up for the loopholes in students'knowledge points,and improves the efficiency of students' learning.Finally,with the curriculum knowledge graph as the knowledge foundation,and the question recommendation method as the main line of the logic layer,a personalized question recommendation system based on the junior middle school mathematics curriculum is designed and implemented.The system includes question exercise module,learning module,knowledge point search module and personal center module and other functions,among which the modules involved in the recommendation model are mainly question exercise module and learning module.The system verifies the practicability of the question recommendation method designed,makes the question recommendation traceable,and improves user satisfaction.