Enhanced Personalized Learning Recommender System Based On Knowledge Graph

With the advent of Big Data Era and the rapid development of information technology,the amount of data is increasing day by day.The education industry is also experiencing informatization and digital innovation,which greatly facilitates people's production and life.At the same time,the development of artificial intelligence has brought new breakthroughs to the education industry.The combination of knowledge graph and recommendation technology can provide students with more accurate and diverse personalized recommendations.The enhanced personalized learning recommender system providing math questions based on knowledge graphs studied in this thesis is proposed based on this background,and mainly includes the following contents:1.According to the basic definitions and theorems of elementary mathematics,a conceptual knowledge graph is constructed,and Trans E is used for training to translate entities,relationships and triples into vector representations,in preparation for the system to implement graph-based personalized recommendation subsequently.It converts each independent and unstructured math question into a structured triple knowledge representation,removing meaningless and redundant information in the question,and constructs a question-instantiated knowledge graph.Then through the means of graph mapping,the independent questions are linked to map the instantiation-concept knowledge graph.2.In order to realize the personalization of users,this thesis creatively proposes the strategies of ”sibling diffusion” and ”multi-hop propagation”.Based on the instantiationconcept knowledge graph,triple knowledge is diffused to help users complete the expansion of interests to achieve a variety of personalized recommendations.At the same time,users' historical data is introduced to control the randomness of multi-hop propagation,and expression evaluation is added to the similarity judgment of questions to realize enhanced personalized learning.A complete set of recommendation architecture process can solve the problem of personalized recommendation of elementary mathematics questions,get rid of the traditional knowledge point labels,analyze the features of the question from the perspective of the content and semantics of the question,and get more accurate and diverse personalized questions.3.In addition,this thesis also proposes and implements a questions recommender system based on the word embedding language model BERT,which studies question's features from the perspective of question's text sequences.Preprocess the question text data in the questions database to remove the variable-length expressions,and use the processed data to fine-tune the BERT Chinese pre-training model to obtain the MATH-BERT model? train the question text into a fixed-length vector representation,through vectors' interactive computing to obtain similar items.Through such means to complete the user's personalized recommendation,it provides another way of thinking for the recommendation of elementary mathematics questions.Finally,this thesis tests the functions of the enhanced personalized learning recommender system based on knowledge graphs through the data of three users,and shows final realization of the functions of ”sibling diffusion”,”multi-hop propagation” and ”enhanced personalized learning”.At the same time,it is compared and analyzed with the test of personalized recommendation system using the language model BERT,which verifies the complete realization of the enhanced personalized learning recommender system based on knowledge graph and the extension function of user's interest preference.