A Group Recommendation System Based On Social Influence

In People's daily social life,attending group activities is an essential part.For a group recommender system,it is very important to suggest pleasurable activities to a group of users.The main challenge in this work is how to learn the social influence of each user,and how to adapt their mutual social influences to different groups,which we can dynamically adjust the mutual item strategy for a group to capture the complex group decision-making process.However,these dynamical strategies are very simple to model the real and complex of group decision making moreover,group members should have occasional influences or burdens in a group,and the burden of a user can vary in different groups.Therefore,an ideal group recommender system should be able to learn user mutual preferences.This thesis proposes a novel recommender system based on mutual social influence for group activity.In particular,the value mechanism for understanding the social implications of individual influences and adapts their shared social impacts to the various association and develop a unique deep social impact gaining knowledge of framework to exploit and combine users' universal and local social community structure facts to enhance the estimation of users' social influences.To overcome the predicament and sparsity of intersection knowledge provided by occasional groups,we applied this proposed model on CAMRa2011 dataset to learn the embedding to make a recommendation from the user-item interaction and user-user interaction.This knowledge learned from the pattern is used to make recommendations of group items to users of the same Group.We adopted a well-known event dataset where we experimented with a real-world data,which is a publicly available dataset published from the context-aware movie recommendations contest.This work achieved a state-of-the-art result of compared to other works.The reported results exceed the other methods with an average of 8.91% and 14.56% for HR and NDCG respectively,for the user recommendation system.Also,this work recorded a convincing result for the group recommendation system;0.68% and 9.93% for HR and NDCG respectively.