Research And Implementation Of Hashtag Recommendation Algorithm Based On Graph Convolutional Network

As one of the unique attributes of social networks,Hashtag plays an important role in information integration,event retrieval and topic participation.However,the low usage rate of Hashtag on social networks makes Hashtag unable to play its role.Therefore,how to recommend appropriate Hashtag to users and improve users’ use of Hashtag has become a hot research issue.The existing Hashtag recommendation research mainly focuses on the Hashtag of single-modal information,and pays less attention to the Hashtag using multi-modal information and user’s historical habit information.For the Hashtag recommendation problem,this thesis focuses on the fusion of multi-modal features and the analysis of users’ historical tagging habits,proposes two recommendation different models,and implements a blog system based on the algorithm model.The specific work is as follows:(1)For the information posted by users in the Twitter social platform,a Hashtag Recommendation Model Based On Self-Attention Mechanism And Graph Convolution is designed.The model uses the attention mechanism to obtain the key information in the picture information.Multiple feature extractors are used to extract text features and fuse them to form new text features.The information of different modalities is used as the node information in the graph structure respectively.Self-Attention is used to obtain the adjacency matrix of the graph structure,and the two-layer graph convolutional network is used to update the node information to complete the fusion of multi-modal features.The commonality and uniqueness of features between different modalities are guaranteed.Various comparative experiments are carried out on the Dataset,and the results show that this model has better recommendation performance than other algorithms.(2)For various user data in the Instagram social platform,this thesis proposes a Hashtag Recommendation Model Based On Graph Convolution And Users’ Historical Habits.The model obtains the historical information published by the user through the user’s ID,and uses the improved Co-Attention network and point product to obtain the semantic similarity between the user’s historical information and the post to be recommended,so as to recommend personalized Hashtags to the user.The experimental results on the public data set verify that the Hashtag recommendation effect of this model is better than other Hashtag recommendation models.(3)This thesis designs and implements a blog system,and applies the model designed in this thesis to the system.Complete the overall architecture design of the system and embed the trained model into the blog.The development and testing of the system are completed.