An Enhancement on Mobile Social Network using Social Link Prediction with Improved Human Trajectory Internet Data Mining

Abstract

Generally, the mobile social network has missing and unauthentic links. The prediction of those links is one of the major problems to understand the relationship between two nodes and recommends the potential links to the users derived from the history of user-link interactions and their contextual information. The recommendation problem can be modeled as prediction of the future links between users. Many research works have been developed to understand the relationship between the nodes and construct the models for missing or suspicious links prediction. Among those, Improved Multi-Context Trajectory Embedding Model with Service Usage Classification Model (IMC-TEM-SUCM) has better enhancement on human trajectory data mining by classifying the internet traffic. However, this method requires the prediction of the relationship between the nodes and social links. Hence in this article, the IMC-TEM-SUCM is proposed with the Social Link Prediction (SLP) mechanism for identifying the relationship between two nodes and predicting the stable links. In this technique, a number of nodal features are considered and their influence on the link prediction problem of Foursquare and Gowalla are examined. This extended network is used for computing two features such as optimism and reputation that depict node’s characteristics in a signed network. After that, meta-path-based features are considered and their influence of the route length on the problem of link prediction is examined. Moreover, a link prediction process is performed by using the machine learning classification algorithms that use the extracted node-based and meta-path-based features. Also, Cosine coefficient and Jaccard coefficient similarity-based techniques are used for computing the similarity index between any two nodes. A higher similarity indicates a higher chance of forming links between them. Finally, the performance effectiveness of the proposed model is evaluated through the experimental results using different real-world datasets.