Research On Social Interest And Personalized Recommendation In Social Networks

With the rapid development of computer and Internet technology,the increasing popularity of smart terminals such as electronic computers and mobile phones has made the prevalence of social networks inevitable.In social networks,the way people communicate and cooperate has changed dramatically,so that people's understanding of themselves and society has been affected.Therefore,in order to make full use of social network data such as users' personal information,interaction data,social relationships,and network topology to mine users' potential interests,conduct community discovery and community evolution research based on interests,and finally achieve personalized recommendations,an interest socialization and personalized recommendation technology in social networks emerges as the times require.At the same time,interest mining,interest community discovery,interest community evolution,and personalized recommendation are all key components of the research on interest socialization and personalized recommendation in social networks.Their performance also has a crucial impact on the overall performance of interest socialization and personalized recommendation technology in social networks.Therefore,the research on interest mining,community discovery,community evolution and personalized recommendation has great theoretical significance and practical value.Firstly,this dissertation systematically studies the relevant theoretical knowledge and key technologies of social networks and identifies some key limitations in the recent existing research works.Then,this dissertation proposes novel approaches to address these limitations focusing on interest mining,community discovery,community evolution and personalized recommendation in social networks.The core contributions of this dissertation are listed as follows:(1)In order to solve the problem that social network data contains short text,irregular language,a lot of noise data and is easily affected by data sparsity and interest drift,which results in lower efficiency and accuracy of interest mining and community discovery,a user interest mining and community discovery model based on trust propagation is proposed in this dissertation.Firstly,a HLDA model is proposed,which combines the improved HITS algorithm and the LDA algorithm,to extract high-quality and popular interests by making use of the inseparable link between interests and their corresponding users.Secondly,in the interest mining,the social information,the interaction data and the user-interest matrix in the social network are fully used to calculate the user trust,the social similarity and the user similarity.The user trust,the social similarity and the user similarity are integrated to mine the potential interests of users more efficiently and accurately.Thirdly,the HITS algorithm and collaborative filtering algorithm are used to improve the label propagation algorithm,and an interest community detection method HLPA is proposed to more accurately and effectively divide the user's community interest.Finally,a comparative experiment with the existing classic algorithm on the data set obtained in the real social network proves the effectiveness of the method proposed in this dissertation.(2)The existing dynamic community discovery algorithms ignore the interaction of information factors,which makes them vulnerable to data sparsity and interest drift.They cannot accurately detect complex and diverse interest community structures in dynamic networks and explore their practical application value.A dynamic user interest community discovery model based on evolutionary clustering is proposed.Firstly,preprocessing methods based on HITS algorithm and S-step extended similarity matrix algorithm are used to obtain high-quality users and posts,optimize the sparse adjacency matrix.Secondly,the mutation algorithm based on label propagation algorithm is used to divide the social network into individual users with different interest community structures,and then through selection,crossover and mutation operations,the best individual user is finally selected.Thirdly,the combination of multi-objective genetic algorithm and label propagation algorithm generates users with specific community structure and enhances the scalability of the algorithm.Finally,a comparative experiment with the existing classic algorithm on the data set obtained in the real social network proves the effectiveness of the method proposed in this dissertation.(3)To solve the problem that existing community evolution algorithms can not achieve ideal results in terms of efficiency,accuracy and scalability in large-scale social networks,a user interest community evolution model based on community topology is proposed in this dissertation.Firstly,the topic scoring method based on the authority value and the minimum distance of the post are used to calculate the number of users who constitute the core sub-graph to ensure that these users have the greatest influence.Secondly,the user interest community evolution model based on subgraph matching is used to accurately detect the corresponding community in the community evolution.And by introducing core subgraphs,it is possible to quickly capture community evolution events,including birth,growth,shrinkage,merger,division,and death.Thirdly,an improved dynamic weight subgraph matching algorithm is used to update the core subgraph in time when the interest of the core user changes.Finally,a comparative experiment with the existing classic algorithm on the data set obtained in the real social network proves that the method proposed in this dissertation has better clustering accuracy,higher operating efficiency and greater scalability.(4)The existing recommendation algorithms are affected by sparse data,cold start and single view recommendation,resulting in too little information for statistical inference,inaccurate understanding of the semantics of post content,and lower recommendation accuracy.A personalized user interest community recommendation model based on multi-view collaboration is proposed in this dissertation.Firstly,the HITS algorithm and the text information mining method based on the description of the post are used to filter out users and posts with high influence,mine the text information of the description of the post and obtain the user's original rating.Secondly,the opinion pre-filtering method is used to comprehensively measure the excavated user's emotional tendency and the original rating level,so as to correct the deviation of the user's original rating from the user's actual interest preference.Thirdly,similarity calculation based on post content and multi-label interest relevance calculation based on mutual information are used to construct a recommendation model based on similarity of post content and a recommendation model based on relevance of post content.Fourthly,the collaborative training strategy is used to achieve the fusion of the three recommended views,and a data selection strategy based on confidence level estimation and cluster analysis is added to the collaborative training to eliminate the distribution bias of the data added to the training data pool in the iteration training and obtain the final recommendation results.Finally,a comparative experiment with the existing classic algorithm on the data set obtained in the real social network proves that the method proposed in this dissertation has a better recommendation effect and can solve the cold start problem to a certain extent.