Personalized News Recommendations In Social Networks

In the digital economy era,with the continuous penetration of big data and social media applications into the lives of network users,the rapid increase in the number of users has promoted the rapid development of recommendation information technology for social networks.Social networking platforms can not only serve as effective channels for users to obtain and disseminate information,but also serve as important media for the latest information and governance of social issues.Providing users with news that suits their personal interests is an important task of social networks,and excellent recommendation algorithms have become the core technology of social network platform competition.However,the recommendation algorithm also has factors that affect its performance,such as ambiguity,sparse data,and crawling a large amount of useless information.Therefore,in order to improve the performance of the recommendation algorithm,the paper studies personalized recommendations for social network data,designs a crawling strategy for social network data sources,and uses text processing and deep neural network technology to carry out research work,as follows:First,the thesis studies the collection and processing of social network data.We designed a web crawler strategy and developed a data crawling method based on the Scrapy framework.This method can quickly customize the collection of target data,and has strong scalability.For the Sina Weibo platform,we collected Sina Weibo data in combination with the designed crawler strategy.In data management,there is a large amount of useless information in Weibo that needs to be removed.The paper uses word segmentation and stop word removal technology to obtain noise-free data,thereby realizing effective data management.Secondly,in the user tag mining algorithm,in view of the ambiguity of user interest tags,the paper uses the Text Rank _TFIDF method to generate keywords and their weights from user microblogs,and combines the Apriori algorithm to generate frequent item sets and association rules.A recommendation algorithm based on extended user interest tags is designed.Experimental results show that the algorithm improves the accuracy of user interest tags.Third,in view of the data sparse problem of collaborative filtering algorithm,the paper uses the latest RJaccard coefficient to calculate the similarity.This coefficient classifies related neighborhoods by considering all the user's rating vectors,and combines it with the User-based Collaborative Filtering recommendation algorithm.A collaborative filtering news recommendation algorithm based on RJaccard coefficient is designed to alleviate the problem of data sparseness and generate recommendations in a lower computing time.Through experimental comparison,it can be seen that this algorithm is better than other classic collaborative filtering recommendation algorithms in accuracy and recall.Fourth,the paper proposes a method for recommending news with mixed features based on deep neural networks.This method first uses an extended user interest labeling algorithm and a collaborative filtering algorithm based on RJaccard coefficient to obtain candidate recommended microblog news,generates a mixed feature vector from this news,and constructs a mixture of user personal interest,behavior similarity between users,and news quality.Features are used to describe candidate news;then the mixed features are input to a deep neural network with multiple hidden layers,and finally the news is sorted through the mixed feature model of the deep neural network to get the final news recommendation list.An experimental analysis of the algorithm is carried out on the Weibo data set obtained by a web crawler.The results show that the algorithm has better results than the five classic personalized recommendation algorithms in terms of accuracy,recall,and F1 measurement.In summary,the paper designs a crawling strategy for social network data sources,improves similarity calculation methods and user interest labeling algorithms,and builds a deep neural network hybrid feature model.Fully extract the effective information of users'personal interests,similarity between users and the quality of Weibo,and further improve the accuracy of personalized recommendations.