Research On Recommendation Algorithm Based On User Relationship In Social E-commerce Platform

With the development of mobile Internet and artificial intelligence,online social activities gradually play a significant role in life.As the combination of other platforms and socialization has become an inevitable trend,the concept of "social network+" has emerged.How to create the personalized recommendation to the e-commerce,media and video platforms based on dynamic characteristics in social networks has become a research focus for Internet applications of next-generation.This paper combines the information of social network with the personalized recommendation of e-commerce,focusing on how to extract effective features as well as improve the recommendation quality in the non-acquaintance network.Based on this background,this paper defines and analyzes the network topology of the social e-commerce.By combining the characteristics of heterogeneous information in social networks,extracting the effective features from the key factors that affects the strength of social relationship among users,the social relationship strength calculation model based on dynamic characteristics is obtained.The feature attributes of this model include the comment features of user,the behavior features of user and the personal information features.On the one hand,according to the shortcomings of traditional topic models in dividing time windows,a dynamic topic model based on automatic time division is proposed.This paper divides the text set non-equidistantly through the similarity of time windows,and then uses the time decay function to correct the user's topic probability distribution,finally uses the cosine similarity to calculate the comment-based user similarity.On the other hand,considering that users usually compare similar commodities before placing an order,the behavior process often includes rich user preferences.Based on this,a feature extraction method based on alternative cost is proposed.The method estimates the substitute commodity in each unit behavior sequence according to the behavior effective function,and establishes an objective function between the purchased commodity and the substitute commodity.By iteratively training the feature vectors,the behavior-based user similarity is obtained.In addition,as a supplement,information of the user such as age,location,and gender is used to draw conclusion on the similarity of personal information.This paper uses the above three similarities to quantitatively assess the strength of social relationship,and then obtains a social relationship strength calculation model that integrates various factors.Due to the huge amount of data in real system,this paper uses the improved k-means algorithm to divide the user into communities.The system first determines the community to which the target user belongs when recommends friends,and then searches for the neighbor users with the closest social relationship in the community to recommend.After verifying on the real e-commerce dataset,the solution mentioned in this paper can improve the average accuracy of the system recommendation and reduce the computational complexity.The solution can also be promoted and applied to other "social network+" platforms such as media and video.