Using Modal Symbolic Data In Group Modeling Techniques For Group Recommenders

With the rapid development of IT technology, Internet has been widely used in people's daily life. Internet brings people with convenience, as well as the problem of “information overload”. People find them have to spend a lot of time and energy to get what they really want from the massive information. Under this circumstance, more and more scholars in IT domain begin to focus on the research of recommender system, which has been proved to be a powerful tool to address the problem of information overload in the Internet. And as a lot of people's activities involve groups, group recommender system, which aims to satisfy the needs groups made up of more than one user, has also attracted plenty of attention in recent years. There has been some recognized research achievements in group recommendation field. But group recommendation is not the simple combination of individual recommendation.it is much more complex and facing more problems than individual recommendation. However, the majority of current group recommender systems do not take these problems into consideration. To increase the accuracy and efficiency of group recommender system much more researches need to be done.In this paper, we present a group recommendation algorithm which makes it possible for tacking the preference of the group as a whole. With the support of symbolic data analysis and genetic algorithm, symbolic model of the group is established. In this way, the whole group can be viewed as a virtual user which enables the method of user-based collaborative filtering to be applied into group recommendation domain. Group modeling is actually an optimization problem. We let the total distance between the models of group members and the whole group as small as possible. We conducted three sets experiments based on Movielens 1M to evaluate the proposed algorithm. The results show that the group models established by the proposed algorithm can reflect the preference differences of group members which means it is reasonable and meaningful to make group recommendation based on these symbolic models. And following the proposed algorithm, we can provide more accurate recommendation results to the group than the comparison algorithms.