A Friend Recommendation Algorithm Introducing Link Prediction Mechanism Based On The Meta-path

With the rapid development of the Internet technology, the fast popularization of mobile terminal equipment,Human-being’s working style,life model and learning patterns have been changed gradually.people have gone from the age starved of information into the era information explosion.The explosion of information is not only huge information content, which presents more on unsuspected variety of information and excessively complex structure of network.Which severely hinder the dissemination of information in the Internet,and bring a new challenge to data mining research:resource consumers don’t know how to start when they face the complex and changeful information resources,while resource providers can hardly effectively grasp the demand and interest of consumers,and give some reasonable and highly available recommendation.some research suggests that recommendation system is a kind of tools and technology to effectively solve information-overload problem.The system presents the useful information to the user,maximize resource utilization,relieve the pressure of information exchanges by fully mining given data,analyzing data relationships and intelligently modeling.The key work of recommendation model is to study the similarity and correlation between the target user and candidates.Link prediction is able to adapt to complex network,reproduce and predict the link that has not been observed or possible connect in the future,by analyzing the the exiting information about relations between nodes.In LBSN,the position information implies a lot of semantic information which can reflect the user’s interest and preference. Therefore, it is an effective improvement and innovation to introduce location information into the recommendation process.Based on the background above,this paper present a link prediction mechanism process the Meta-path to recommend online friends.Introducing the Meta-path to describe the relationship between the types of nodes in the heterogeneous networks.Firstly,working out the numeric of all Meta-path based on random walk algorithm.After that,getting theweight of Meta-path by maximum likelihood estimation,and the connection probability of given node pairs.Finally,adopt a logistic regression model to evaluate the optimum fitting, generates the list of suggestions.In the fourth part,the simulation result concluded that,in terms of precision rate and recall rate, the algorithm is better than the other two algorithms.Compared with merely based on position score,precision rate is increased by 10% wihle recall rate 13%improved. As compared with mutual friends, precision is improved by 14% while recall rate 12% increased.Which indicates that this algorithm significantly improve the effect of the recommend,and that the checkins has a considerable potential value as well as research to the recommended application.