Research On Resource Recommendation Method Based On Machine Learning In Social Network

With the development of Internet technology and the popularity of Internet users,the number of users exploded and the growth of social network services are also undergoing dramatic changes.In this case,the user wants to quickly get the information of interest or content providers want to recommend targeted,need an effective social network recommendation technology.But the traditional recommendation technology only consider the user's scoring of the project,the project's own attributes or history browsing records,and does not take into account the changes in user interest laws,nor adapt to new social networking features,such as access to user data,user interest Diversification and so on.In addition,the size of social network data sets are often huge,in which case the traditional recommendation of the efficiency of technology is relatively low.To solve the above problems,this paper studies the recommendation method based on machine learning in social network,and synthesizes the related technologies of user clustering,prediction and recommendation,and establishes the corresponding recommendation system.The system uses Sina microblogging user data,preprocessed to extract the data characteristics,and then use the clustering method based on Word2vec to get the clustering results,then the Markov chain is more interested in forecasting,and finally based on the prediction results to the user recommendation.The system takes into account the fact that the change of interest in real life has a predictable regularity and the diversity of users' interest to build a model to provide users with interest recommendation.The experimental results show that the proposed system has good performance.In this paper,the recommendation method based on machine learning in social network mainly includes the following contents:(1)Analysis of micro-blog data in social network,the basic natural language processing,including the use of NLPIR Chinese word segmentation system for word,remove stop words,synonyms replacement.Taking into account the user's micro-Bo published text data is generally longer,need to extract the main features for further processing.The system uses the Word2vec model to process the data to get the corresponding user characteristic vector.(2)Clustering algorithm is used to aggregate the users with similar attributes into one class,and the users are more interested in the prediction.This paper proposes a clustering method based on Word2vec to achieve clustering,which can narrow the searching range,improve the efficiency and make the users locate more accurately.Then,considering the diversity of users' interest,a multi-Markov chain multi-interest forecasting model is proposed,which can predict the user's interest according to the user's characteristics and build the user's demand characteristics.(3)Based on the above results,this paper proposes a keyword filter based on the traditional user-based collaborative filtering recommendation algorithm,which combines the historical interest and the forecasting interest into the user's interest.The proposed system takes into account the attributes of the user's multi-interest category,and uses two machine learning algorithms,clustering and prediction,to improve the performance of the recommendation.The experimental results show that the proposed method has good performance.