Research On Multiple Recommendation Algorithms For Event-based Social Network

Event-based social network(EBSN)provides users with a convenient and efficient entertainment platform where they can launch or organize some offline events by using online social networks.Event participants can share their feelings,choose interest type tags,set up common interest group and organize events online.Therefore,it is of great practical significance to study the interaction between users’ social relationships in virtual space and their social events in physical space.There are huge application prospects to design recommendation algorithms to promote user’s experience.The recommendation in EBSN mainly refers to recommending group,event,tag to user and recommending tag to group.It is a kind of multi-recommendation problem.Event-to-user recommendation is a special cold-start problem.Firstly,the holding time of event is a finite and transient time slot in the future.Secondly,the effective recommending time of event is a finite time slot from release time to holding time.The current research work mostly regards multi-recommendation as kinds of single recommendations respectively.The multiple properties and relationships of heterogeneous nodes(user,event,tag,group)in EBSNs are not used enough.More efficient implicit feature representations are not mined.We propose multi-element combination method based on machine learning technology and deep learning technology.We use link relationships between nodes to deeply mine latent factor in order to make multi-recommendation.We combine latent factor with direct property to solve cold-start problem and improve recommended result.The main contributions of this paper are as follows.(1)We propose multi-element combination method based on machine learning.We utilize link relationships between heterogeneous nodes to make multi-recommendation.The means of multi-element combination contains of three respects.1)The multiple links between various nodes are utilized to make deep learning and mine implicit feature representations of each kind node.2)The machine learning model is constructed by merging Matrix Factor decomposition,Bayesian Individualized Ranking and Artificial Neural Network algorithm.3)We design twelve kinds of strategies from three perspectives which are objective function,combination way of vectors and optimization way.This method can not only mine more features but also can reduce noise during train.We make experiments with Meetup datasets and verify the effect of our method.(2)We respectively use content-based collaborative filtering algorithm and attribute mapping method to solve cold-start problem in EBSN.Then we make improvement based on them.1)We make multiple combinations of direct property weights and explore the best way of weight combination.2)We propose multi-attribute mapping method to extend attribute method.We convert diversity relationships of social links to link-properties of target node.They are combined with direct properties to get whole explicit features.The mapping idea is used to get implicit features of new event.We combine the implicit features and explicit features to make event recommendation.We make experiments with Meetup datasets and verify the improvement of our method.