Research And Application Of Personalized Recommendation Algorithm Based On Network Representation

Nowadays,with the rapid development of the Internet,users can search any information they need and buy the goods they need through the Internet.However,in the era of e-commerce,users tend to use social networking sites to browse or spread all kinds of information,resulting in the rapid increase of network data.In this case,recommendation system followed and gradually developed.Recommendation algorithm is the most important part of a recommendation system.Although the application of traditional methods has a good recommendation effect,there are still some shortcomings.At present,personalized recommendation algorithm is mainly based on static network,which can not accurately model dynamic network and carry out effective recommendation.At the same time,there are some problems such as sparse data,low recommendation accuracy and low recommendation diversity.This paper puts forward the corresponding solutions to the above problems,and the specific research is as follows:(1)In order to extract dynamic information and multi-layer network features of dynamic network over time,a personalized Tner recommendation algorithm based on dynamic network representation was proposed.Time coefficients were added into the loss function,and different time dimensions were used to adjust the loss function for optimization training.At the same time,according to the change of nodes,the number of hidden layers is extended to deal with the problem of nodes changing with time in dynamic network.The results on multiple time series data sets show that Tner is superior to other algorithms in accuracy rate,recall rate and F1-Score,which reflects the recommendation accuracy and efficiency of Tner algorithm.(2)Aiming at the problems of low accuracy and poor diversity of traditional methods,a weighted two-part network recommendation algorithm based on energy distribution was proposed.The low frequency item threshold is proposed to delete the low frequency items in different proportions of data,and the energy distribution is combined with the weight formula to assign corresponding energy to the recommendations between user items.The improved weight formula was used to calculate the similarity and make recommendations.Finally,logistic regression method is proposed to modify the recommendation list to make the final recommendation list more accurate.The results on two datasets show that the recommendation accuracy and diversity are higher than existing algorithms.(3)Finally,combined with the above two recommendation algorithms based on network representation,a recommendation system based on network representation is designed and implemented to realize personalized film recommendation.The system integrates the functions of user account registration,login,query of film details,film scoring and film recommendation.