Research On Group Recommendation Method Based On Deep Learning

In the boom of artificial intelligence,group recommendation has made great contributions in various fields,and has become a continuous hot research issue in the field of recommendation.There are obvious differences between group recommendation and personalized recommendation.In terms of the recommended audience,group recommendation takes the group as the audience,which is in line with the current situation that people's group activities are gradually increasing.How to better solve the preference conflicts among group members and form more accurate and diversified group preferences has become the main task of group recommendation.Therefore,the maturing deep learning technology has also brought new vitality to the research of group recommendation.Based on group recommended method carried on the thorough research and analysis on the basis of the group around the current recommended performance problems,auxiliary information utilization problem,users tend to move sex preference and interaction,combining with the characteristics of deep learning technology,from two aspects of group found fusion and preferences,launched on the research of the group's recommendations,given a group recommended based on the technology of deep learning model,implements the group recommended results of optimization and performance of ascension,and the validity of the model is verified by experiment.The specific work is as follows:(1)To solve the problem of low intrinsic similarity of group members caused by time-transience of group users' interest tendency in group discovery,a group discovery method based on short and long time memory network is proposed.Group found methods combined in the method of implicit levels of access to information and applications,through the length of the hierarchy of implicit memory network information mining,fusion of users tended to explicitly,the user implicit,project information such as the relationship between implicit,defines the general tendency to computational method for the user's preferences,and calculated according to the users tend to generate high similarity,hidden level information and effective use of high quality group found that and the effectiveness of the proposed method is verified by experiment.(2)Aiming at the inaccuracy of preference fusion result caused by the change of preference preference caused by user interaction behavior in preference fusion,a preference fusion method based on convolutional neural network is proposed.The method using user behavior data to construct the history of the user interaction model,the interaction of the weight matrix is obtained by convolution neural network to feature extraction,get effective user interaction affect weight,and according to the weight weighted score matrix to the customer,on the basis of grading recommended preference fusion method combined with recommended fusion,the optimization of the preference of fusion,and the effectiveness of the proposed method is verified by experiment.(3)Combined with the proposed group discovery method based on short and long time memory network and preference fusion method based on convolutional neural network,a group recommendation model based on deep learning is constructed.The model using two different deep learning technology in groups recommended for application integration step by step,implements both group of users to explore the relationship between the implicit,and according to the different members interaction influence degree computing group preference,solved the group recommended by group members in the process of low similarity and preferences within fusion of subdivision recommend effect difference problem,achieved to improve the performance of recommendation accuracy and optimization recommendations.