Research On Group Recommendation Based On Deep Learning

In recent years,due to the vigorous promotion of social media,online group activities have become more frequent.For this reason,group recommendations have gradually entered the sight of the researcher.On the other hand,with the great success of deep learning,research on deep learning in the field of recommendation systems have become more popular.In this context,this article under the multi-task learning framework to attempt to strengthen user group recommendations by using attention networks to dynamically capture user preferences,and to strengthen group recommendations in social networks through multi-level attention networks,and enhanced item list recommendations through self-attention networks and gated networks.Therefore,the main work of this article includes:1.Due to the current group recommendation systems almost adopt a predefined strategy to aggregate the preferences of users in the group.This paper proposes a recommendation model,namely AGREE,based on attention mechanism.Specifically,this model can adjusts a group's preference aggregation strategy dynamically through an attention network and simulates the complex process of group decision-making.Subsequently,in order to be able to learn the complex interaction information between users,groups,and items,we use NCF models to simultaneously model user-item and group-item interactions under the multi-task learning.This approach makes group recommendation and user recommendation can be mutually enhanced to achieve better performances.To validate the effectiveness of AGREE,we performed extensive experiments on two real-world datasets.The results show that AGREE achieves state-of-the-art performance for group recommendation.2.Due to the current group recommendation model does not make full use of the shortcomings of user social network relationships.Based on the AGREE model,this paper proposes a multi-level attention network model So AGREE.In this model,The first level of attention network models user-follower interaction information to integrate social information into the user representation.Then,another attention network is applied to model user-group interaction to learn the group representation.Through this multi-level attention network,the model can learn the embedding of users and groups.Finally,through the multitask learning mode,a neural collaborative filtering model with shared parameters is adopted to simultaneously learn the group recommendation task and the user recommendation task.The experimental results show that the model proposed in this paper has achieved better group recommendation results after adding user social network relationships.3.Due to the shortcomings of the current item list recommendation model that only obtains user preferences at the item level,this paper proposes a list recommendation model GANCF based on attention networks and gated networks.Specifically,this model employs the attention network and the self-attention network to learn the user's preferences for the items and attributes in the list,and applies a gated fusion layer to fuse multi-level preference to obtain an embedding representation of the list.Finally,through the multi-task learning mode,the model inputs list embedding,user embedding,and item embedding into an outer product-based interaction network to learn user-item and user-list interaction information simultaneously.Experimental results show that the model in this paper can obtain good list recommendation results.