The Design And Implementation Of Mind Map Recommender System Based On Dual-Channel

With the arrival of the information age,the explosive growth of data size has caused users to face problems such as data overload,making it difficult for users to find and discover the content they need.Mind maps,as a learning tool,are increasingly being known and used by users,but its recommendation platform is rarely known,hindering the spread of mind maps.This system starts from the characteristics and needs of the mind map recommendation platform and provides accurate recommendations for users.(1)This paper proposes a feedback-based graph convolutional network(FBGCN)that provides more accurate recommendations leveraging implicit and explicit feedback.In order to combine the two feedbacks,a global explicit feedback module based on the pairwise learning-to-rank method is proposed.After experimental validation,this model has shown a 15%performance improvement compared to the LGCN model.(2)This paper proposes a session recommendation model based on user interest transfer(ISB-GCN),aiming to provide accurate recommendation results for anonymous users accessing the system.The model constructs an interest decay graph based on the forgetting curve,from which it extracts the user’s global and local interest features.The model also constructs an interest container to capture the user’s interest transfer characteristics.Experimental validation shows that this model has a 10%performance improvement compared to the GCE-GNN model.(3)A sentence embedding representation learning model called EqvCSE based on equivalence contrastive learning has been proposed,which generates sentence embeddings used for topic similarity recommendation.The model achieves insensitive data augmentation by inputting the sentence to the encoder twice,and a discriminator is designed for sensitive data augmentation such as word replacement,which improves the performance of sentence embeddings.Based on extensive experiments,the model outperforms the SimCSE model by about 5%,reaching the SOTA level.(4)Based on the three models mentioned above,this system has been designed and implemented as a recommendation system for mind maps.The system adopts a dual-channel architecture of personalized recommendation and session recommendation,covering recommendation tasks in both login and anonymous scenarios,thus meeting the diverse needs of users for mind map platforms.