Trust And Similarly Labeled Link Prediction Algorithm

Link prediction based on Social networks is a research direction of data mining,Social networking is composition of the social entity and the relationship of them. Thepurpose of link prediction is to find those hidden relationships by explicitrelationships. Traditional data mining is usually carried out research on the basis ofthe entity properties information, and the link prediction research on the basis of theentity relationship. Application of link prediction very wide, In the field of Sociology,It can analyze the social evolution process; In electronic information field, Linkprediction can be used to create various recommendation systems; In the fieldof biological information, Link prediction can be used to analyze the interactionbetween proteins. Therefore, it has a very important value to carry out linkprediction research on social networks.At present, most of the link prediction algorithm is studied from thefollowing two aspects:1,Similarity Perspective, using the similarity between entitiesto predict the hidden link, and the similarity function is designed by topologyinformation of social networks. However, the existing link prediction algorithm isnot fully used the network topology information, such as Common Neighbor, It onlyuses the number of common neighbors between entities as similarity index，Whileignoring the mutual relations between the neighbors;2, Trust Perspective, Ituses between the real social network topology information and attribute informationfor link prediction, However, during the calculation of trust degree, every entity ofthese algorithms are always treated equally in social network, and some oftrust computing method has certain subjectivity, such as TidalTrust algorithm,it requires the user to score their direct neighbors before calculating trust degree. Inaddition, in binary trust networks, trust only two values(0stand for no trust,1stand fortrust), So it can not reflect the real relationship between entities. This paper carriedout the following studies aiming at the above problems:Firstly, the most similarity algorithm does not fully consider the networktopology information, the paper design a calculation method of similarity, it takes intoaccount the property relations between network nodes, and we introduce cosine similarity method in the calculation method, in consideringthe network node attribute relation, while do not ignore the actual textsimilarity between attributes.Secondly, most of trust calculation method is always treat all the nodesequally in the network, while ignoring the itself characteristics of each network node,and some trust computation method has certain subjectivity, This paper proposes anew method of calculating trust, It takes the interaction count between nodes asthe trust weights between nodes, it can better reflect relationship between nodes innetwork’.Thirdly, this paper presents a link prediction algorithm based on trustand similar tag (TAST), most of link prediction algorithms only process the node withthe target tag, and TAST process all the nodes, it can improve the linkprediction’s coverage.