Research And Application Of Educational Resources Recommendation Strategy Based On Heterogeneous Information Network

With the development of educational informatization,the number of educational resources available to students is gradually increasing,and multi-dimensional education,online learning,and other learning methods are becoming more and more popular.Nowadays,students are finding it challenging to swiftly locate learning resources that fit their individual needs in the face of huge learning resources.As a result,experts in the field of education have begun to employ customized recommendation technology based on artificial intelligence to accomplish the goal of "teaching students according to their aptitude." in the era of intelligent education.Education recommendation system can fully analyze students’ historical data,obtain students’ implicit learning preferences,and recommend learning resources that may meet their personalized needs to students.According to the analysis of existing research,the previous recommendation methods only consider the similarity of resource selection between students or recommend learning resources by looking for similar learners,ignoring the impact of other auxiliary information other than students and learning resources on the recommendation results,The heterogeneous information network can exactly model multi-type data and accurately reflect the multi-type relationship between entities,but the multi-type nodes in the network have various node embedded information in different contexts,which has also been ignored by some researchers.Therefore,this study investigates customized recommendations of educational materials based on heterogeneous information networks in order to address the shortcomings of existing approaches,increase the accuracy of personalized recommendations,and suit students’ specific learning goals.This thesis’ s major research work are as follows:(1)Firstly,to consider the various types of auxiliary information contained in educational data,this thesis proposes a high-order preference propagation recommendation algorithm based on a Heterogeneous Knowledge Graph(HKGR).By modeling the library borrowing data set of a university library through knowledge graph(KG),the inherent attributes of book nodes such as author and category are also regarded as entities in the network.Taking a book in the student’s history as the starting point,this method obtains high-order neighbor information from the outer layer along the relationship link defined by the KG,iteratively updates the embedded representation of the starting point at each level through an information aggregation calculation method defined in this thesis,and then calculates the posterior probability of the entity representation based on the current KG through the Bayes formula to get a more explanatory prediction of students’ clicks on books.By comparing the experimental results of some similar models,HKGR outperforms other models in the accuracy of recommendation,which proves the effectiveness of this model and the necessity of modeling multiple types of auxiliary information in the recommendation system.(2)A MOOC knowledge concept recommendation algorithm based on a Multi-aspect Heterogeneous Information Network(Multi-HIN)is proposed,which aims to explore the knowledge concepts that students need to master.Some previous works mainly rely on a single type of node to generate embedded representation,so most of them only have limited presentation ability.Given this,Multi-HIN can consider the impact of various types of entities on students’ learning preferences.To learn more accurate node embedding,the model regards different aspects as students’ multi-dimensional interests.At the same time,the Gumbel-softmax method is used to perform the aspect selection process and dynamically assign an aspect embedding to each node.Through conducting extensive experiments on the public data set MOOCCube,the superiority of this method in the research of knowledge concept recommendation is finally proved.