Social Network Friends Recommendation Algorithm Research On The Basis Of User Behaviors

With the rapid development of Internet technology,the mode of making friends through the social networks has become more and more popular.In order to allow users to make friends,social platforms often provide users with dating channels in the form of recommendations.The current friend recommendation model only focuses on the network structure or user interest,and is relatively single.Moreover,due to the lack of feature information of the network users and the complicated relationship,the network platform filters the information of the user incompletely and the recommendation accuracy rate is low.Therefore,in order to perfect friend recommendation technology,this article has conducted in-depth analysis and research.How to excavate the user's potential features,interactions,and link information,improve the accuracy of friends recommended and other issues.This paper starts with the research of user behavior and discusses the following two aspects:1.Using the machine learning XGBoost model to classify user behavior features,an XGBoost classification prediction model that fuses user features is proposed.By collecting and quantifying information such as user social networking,personal profiles,etc.,a new user feature model is constructed.Through XGBoost iterative iteration to get the user's feature importance coefficient,the feature weight calculation method between users in the recommendation process has been optimized to improve the accuracy of the recommendation prediction.Contrast traditional machine learning algorithms with high accuracy and speed.2.Apply the community discovery model to the friend recommendation technology,and propose a friend recommendation algorithm based on community discovery.Firstly,Doc2Vec is used to do user text mining,and the user's feature is used to build a user's comprehensive topic similarity model.Thus,a user-comprehend similarity modularity function is introduced to generate a user community,and a traditional community-based algorithm based on modularity is improved.By locating the community where the target user is located and calculating the similarity and link strength relationship between the users again,it is possible to recommend the user with the highest similarity.Compared with the traditional recommendation method,this method has a certain improvement in accuracy.This paper classifies user features through the application of machine learning algorithms,uses a deep learning model for text processing,constructs a comprehensive user topic similarity model,combines community discovery ideas with improved recommendation algorithms,and uses microblogging data experiments to successfully implement user friend recommendation.