Similarity Index For Pocket Switched Network Based On Behavior Patterns

The similarity computing between samples has significant meaning and wide usage in link prediction,node clustering,recommendation system and community mining.For example,the link prediction in traditional network based on node similarity has great prediction accuracy with low computational complexity.But in PSNs(Pocket Switched Networks),since it's a kind of DTNs(Delay Tolerant Network)which transmits message based on node movement,the topology may changes frequently.So,the node similarity index based on local and global network information can't achieve expected outcome for link prediction.At the same time,the node movement and data exchanging in PSNs have strongly sociality and regularity because the carrier of node in PSNs is human being.The Studies in this thesis will explain the social characteristic and movement regulation based on move pattern of node carriers,and separate the node similarity into social similarity and movement similarity.On the one side,this thesis get the relational weight based on the times of communications and the duration of each communication.Then gets the social similarity by the relational weight and traditional similarity index like AA or CN.On the other side,it defines the movement similarity based on node frequently visiting areas which based on cosine similarity.At last,the thesis bring an node similarity index which combines the social similarity and movement similarity to provide research reference for link prediction and routing strategy in PSNs.This thesis select MIT Reality and Dartmouth/Campus as experiment datasets.It uses AUC which is the area of ROC(Receiver Operating Characteristic Curve)and Precision as evaluation indicators;groups of contrast experiment are designed for determining the weight between social and behavior similarity.The experiment results show that the proposed similarity based node move pattern(SNMP)has better prediction results and can reflect the real link status between nodes in PSNs.