Group Recommendation Algorithm Research Considering Time Context

With the rapid development of information technology and Internet technology,large amount of information comes out constantly.This brings convenience to people and at the same time,makes people more and more difficult to find the information they need from mass information.Personalized recommendation system is one of the tools to solve the problem of information overload effectively,and its effectiveness can relieve the existing situation.It recommends users with items or information they may be interested in according to the preference and purchase history of users.But obviously,individual recommendation aims at individual users and can not exert effectiveness on group users who want to go out together.So group recommendation has received more and more concern and many people start to devote themselves to studying group recommendation system.Group recommender systems study how to select items from huge amount of information to satisfy group users' needs.As it has proved to be a powerful way to solve the selection problem for social group activities,it does have excellent application prospects.Context information such as the time users get access to websites,location,or mood plays a significant role in improving recommendation quality.But previous group recommendation research is insufficient because of ignoring the importance of uses' context information.Among the context information,time context is the most important one which has deep and extensive influence on user interests,because the preferences of users change all the time,the preferences for items have seasonal effects and items have life cycles.If we want to predict groups' interests more accurately,we should pay more attention to users' recent behaviors instead of treating every behavior fairly.In this study,under the consideration of time context and regarding it as a very significant factor,item model,user model and group model are established using modal symbolic data analysis.Group ratings can not be acquired directly in most circumstances that we define functions to get the group ratings.In this algorithm,we assume the items that were rated longer before have the less influence on user preferences and different time weights are given for different rating time for user and group ratings.Hence,a new group recommendation algorithm is put forward.Then recommendation list is generated based on nearest user neighbors that are found through a dissimilarity function adapted from modal symbolic data analysis domain.The experimental results using movielens show that our recommending algorithm can properly reflect groups' interests and produce high-quality recommendations for larger size groups,both in accuracy and coverage.