Research On Cold-start Social Network Recommendation Based On Autoencoder

In the era of information flooding,it has become more and more common to use recommendation systems to select information of interest for oneself.The recommendation system can predict unknown interaction relationships by using the existing historical interaction data,but the traditional recommendation system algorithm has some shortcomings that restrict the improvement of recommendation accuracy,that is,the traditional algorithm generally cannot solve the data well.The problem of sparseness,and is greatly affected by the cold start problem.In view of these problems,it is a good solution to integrate social relationship recommendations.This social recommendation algorithm can effectively use the user's own friend relationship to make up for the user's lack of project interaction history.However,social recommendation also faces many new challenges,such as the lack of social relationships,the influence of noisy users,and so on.This paper takes the traditional social recommendation algorithm as a starting point,and proposes a social recommendation algorithm based on an autoencoder.First of all,use historical interaction data in social networks to enhance the social relationships of users in the social network topology.Friendship in social networks is very helpful in improving the accuracy of recommendation,but not all friends are valid friends,and Among non-friend users,there will also be potential friends who have a close relationship with the target user.By enhancing the user's social network relationship,the user's friend information can be made more accurate.Imitate the user relationship topological structure diagram to build the project relationship topological structure diagram,and establish similar relationships between the projects based on the historical data of the interaction between the user projects.At the same time,the scoring matrix is used to represent the characteristics of users and items.The feature representations of users and items are expressed in the form of a scoring matrix,which makes it easier to extract feature preferences.However,the scoring matrix is greatly affected by noisy users.In order to reduce the influence of noisy users,this article introduces a self-encoding network to remove noise.Finally,in order to deal with the cold start problem in social network recommendation,this article introduces the concept of popular users based on the principle of popular recommendation.It can effectively alleviate the user's cold start problem by filling in popular users for users to enhance the user's characteristic expression.This paper proposes a social recommendation algorithm model based on selfencoding network,which alleviates the problem of noisy users and cold start in social recommendation.The model has been experimented on real data sets for many times,and compared with traditional mainstream algorithms,it is found that the overall recommendation effect has achieved the best results on the MAE and RMSE indicators.At the same time,better results can be achieved on the cold-start dedicated data set.