Research On Recommendation Systems Based On Link Prediction In Social Networks

As one of the fields of data mining,link prediction has been widely studied in the field of complex networks.With the rapid development of online social networks,the recommendation of links between people has become a key function of social network services.Link prediction is based on the current network structure and the attributes of nodes to predict which users who have not yet made friends "tend to be friends",and send this result to users as "friend recommendation".Therefore,the chain prediction on social networks has practical significance.Compared with traditional complex networks,social networks can have more information to use,such as community characteristics,text information and so on.A good link prediction method often needs some good network characteristics or growth mechanism to support,such as community structure,preferred connection and weak connection effect,which can guide link prediction very well.In this paper,firstly,we propose a mechanism based on relationship strength according to the characteristics of social networks,and improve the existing link prediction algorithm based on this mechanism.Secondly,considering that the network structure attribute and interest preference of users are the important factors that affect the link,further improvement is made by combining the community discovery algorithm,and the relationship information and interest preference characteristics of users are also used.Finally,how to make good use of these network information and mechanism is the key to affect the accuracy of link prediction.Therefore,this paper designs a friend recommendation system which combines link prediction and the label propagation community partition algorithm.The specific research contents are as follows:(1)This paper proposes an improved link prediction algorithm(Combining Node degree and Relationship Strength,CNRS),while traditional link prediction only uses the structural information of nodes and networks and ignores the reasons of the relationship formation on social networks.Firstly,we find the unique characteristics of social networks relative to complex networks.The generation of social network links is based on a variety of relationships.In this paper,the concept of relationship strength is indirectly measured by the common neighborhood compactness;Secondly,the local link index is improved according to the relationship strength,and the calculation method is explained in detail.Experiments show that the nodes with higher relationship strength are more likely to have links in the future,and the more obviously the social network belongs to,the higher the average degree,the more significant the performance improvement.(2)Preference connection has been proved to be an idea that can improve the accuracy of link prediction.On this basis,this paper proposes a combination of the label propagation and the link prediction algorithm.Firstly,we collect the attribute features and text information of users to explore their potential preferences and extract tags,and then construct the user feature vector model to calculate the similarity between users;Then,based on the improved multi label propagation community discovery algorithm(Multi-Label Propagation Algorithm,MLPA),similar communities are mined;Finally,on the basis of community,we use link prediction to find out the node pairs with the closest relationship strength,and select Top-k potential friend list to recommend to users.This method not only improves the accuracy,but also reduces the calculation scale of link prediction.The evaluation is carried out based on the real data set.The experimental results show that our algorithm has achieved better performance than the state-of-art local index method.(3)A social network recommendation system integrating CNRS algorithm and MPLA algorithm is designed and implemented.In view of the problems in practical application,this paper fully considers the data acquisition and preprocessing,and then integrates the recommendation module.Considering that big data technology has been applied to massive social data processing,the recommendation algorithm in this dissertation uses distributed computing,so it has a certain practical value.