Course Recommendation Based On Meta-Path In Heterogeneous Information Networks

With the rapid development of the Internet and the growing demand for education,largescale open online courses came into being in 2012.With the rapid development of large-scale online course platforms,there are more and more courses on online course platforms,and there is a lot of redundancy in teaching content and information overload problems.At the same time,the diversity and uncertainty of information on online course platforms makes heterogeneous information without clear logical relationships and structured organisation,making it difficult for learners to find the knowledge they need in them,thus hindering the online learning This hinders the development of online learning.Traditional online course platform-based recommendations are based on click-through rates,Top-N rankings and user interactions,which are all based on historical user interaction data.However,these methods do not fully consider the possibility of integrating other auxiliary information,ignore the structure and characteristics of heterogeneous information networks,and cannot effectively use information in different forms and from different sources.To this end,this paper proposes a meta-path-based course recommendation system in heterogeneous information networks,which improves the accuracy and personalisation of course recommendation systems by exploiting multiple types of nodes and relationships between nodes and fusing richer semantic information to produce high-quality final embeddings and improve the performance of recommendation algorithms.The main research in this paper is as follows:(1)A meta-path-based course recommendation method in heterogeneous information networks.This paper firstly investigates the heterogeneity and complexity of entities and relationships in online course platforms,identifies node relationships in online course heterogeneous information networks and constructs a heterogeneous information network model for online courses;secondly,uses a random wandering strategy for meta-paths to generate node sequences and uses a skig-gram model to learn the embedding representation of entities in different meta-paths;then,further uses a two-level attention The next step is further vector fusion using a graph convolutional network with two levels of attention,where neighbour-level attention aggregates different classes of neighbours on the target node and meta-relationship-level attention aggregates the semantic contributions of different metarelations and adaptively learns the importance level between different neighbours and relations.Finally,a series of experiments were conducted on the real dataset MoocCubeX,and compared with the optimal benchmark method W-Meta Path2 Vec,the optimal performance was achieved in all metrics,verifying the effectiveness and rationality of the method.(2)Online course recommendation system platform.For the online course recommendation system,four major functional modules are designed for implementation.Firstly,in the data acquisition and data pre-processing module,a crawler is used to crawl the MOOC platform data and use the Hive data warehouse for data pre-processing;secondly,in the data storage module,the relational database PostgreSQL and HDFS tools are used for data access and transformation;then,in the course recommendation module,the above data is used to place in the heterogeneous information network based on the meta-path Finally,in the course service module,the system implements user registration and login,course search and search,and course display functions,and realises user communication and interaction in course learning.The system meets the basic requirements of the online course platform,while providing personalised course recommendations to users through the system’s recommendation function,helping users to choose high-quality courses and avoiding information disorientation in the face of a large amount of course information.