A Group Recommendation Algorithm Based On The Influence Of Group Members

With the development of high-speed network technologies,personalized recommenders is convenient and effective for users to solve the data redundancy.The most of the recommendation services is recommending items to the individual,however,there are some services,like restaurant or movie theatre,which needs to recommend items to a group of users.In this paper,a new method for calculating group similarity to make group recommendations is proposed,based on the existing problems of group recommenders and the new group issue.First of all,due to the existing group recommendation algorithms need group members' personalized parameters,when they take the group members' influence into account.What's more,some of them even need to get their basic information through questionaires or other forms,which exposed their privacy and increased the burden of users;And,in order to aggregate or infer the whole group preference,they need experts knowledge,which makes the recommendation easily affected by the subjective effect.To solve these problems,this paper presents a group members' weight analysis algorithm with consideration of interaction among group members.Firstly,clustering algorithm is used to cluster items which the group has rated and then mining its preference by calculating the score of each cluster.Secondly,using classification algorithms to classify the items which its members have rated and mining every member's preference by calculate the score of each classification.Lastly,the genetic algorithm is used to calculate the optimum solution,which is the influence weight of the group members.Secondly,most of the existing group recommenders need the group rating of the target group or its subgroup to infer group preference or the interaction among group members,which can not solve the case where the target group does not have the group rating.Aiming at this problem,this paper proposes a new group similarity calculation method to solve the new group problem and to improve the recommendation accuracy to a certain extent.Firstly,matching the user-pairs between the general group and the target group,and then calculate the weighted sum of the users' similarity and their influence weight,which is group similarity.Finally,predict the ratings of candidate items based on collaborative filtering and make the TOP-N recommendation list.Finally,The algorithm proposed in this paper is validated and analyzed on the group simulation dataset,and compared with the existing algorithms to validate the availability of our proposed approach.