| With the rapid development of "Internet + education",various education platforms have emerged one after another.Learners can break through the limitations of time and space,learn anytime and anywhere.However,in recent years,with the explosive growth of learning resources on the internet,learners cannot quickly find the right learning resources from the massive data,which reduces the learning efficiency.Therefore,the application of learning resource recommendation algorithm provides convenience for learners to learn online.Based on the existing research results,aiming at the limitations of a single recommendation algorithm,this thesis conducts an in-depth study on video course resource recommendation and text book resource recommendation through the hybrid neural network model.The main work is as follows:(1)As the current video recommendation algorithm only considers the historical behavior data of users and ignores the content preference information of user,in addition inactive users may have a negative impact on collaborative filtering,which leads to problems with recommendation algorithm such as excessive fuzzy recommendation content and data sparsity,a hybrid recommendation model based on adaptive graph convolution and weighted content similarity calculation is proposed.Firstly,the channel attention mechanism is used.By weight useful information is highlighted and useless information is suppressed.Secondly,the self-attention mechanism is added to the layer combination module to further extract the key information and the inactive users’ high-order information.The two kinds of information are disseminated on the user-course interaction diagram to learn the user and course embedding,and the embedding of each layer is combined as the final embedding.Finally,combined with the weighted content similarity calculation,the user’s content preference is mined,and the results of the fusion calculation are recommended.(2)As the current text recommendation algorithm only considers the sequential correlation between historical interaction items,without considering the time variation and diversity of users’ preferences,a hybrid learning resource recommendation model combining gated neural network and self-attention routing algorithm is proposed.Firstly,the cyclic neural network is used to calculate the user embedding according to the user’s historical behavior and the gated neural network is used to extract the preference vector of users and similar users.Secondly,the learner difference factor is introduced and the relative future sequence of each neighbor user is selected from multiple interactive data by filtering through the neighbor extractor,so as to construct the neighbor with the same preference for learners.Thirdly,the self-attention routing algorithm is introduced to soft cluster the historical behavior data of each set into multiple vectors to represent the diversified interests of users.Finally,the attention mechanism based on time perception is used to correlate the acquired future preference sequence with learners’ historical behavior,generate dynamic user preference representation vector,and make diversified recommendations according to learners’ preferences for different items.Experimental results show that the gated neural network and self-attention routing algorithm can effectively improve the performance of the model.(3)A learning resource recommendation system is designed and implemented by using the recommendation model proposed in this thesis.The system can better let users experience the charm of the learning resource recommendation algorithm.Firstly,the thesis analyzes user requirements,designs the system architecture and determines the functions of the system.Secondly,all functional modules are displayed and learning resources are recommended in the form of interface.Finally,the system is tested.The system runs stably,easy to operate and has a beautiful and generous interface. |
