A Research Of Group Recommendation Algorithm Based On Negotiation

With the booming of information technology, information overloading nowbecomes one of the most urgent issues. As the traditional information filtering methodcannot satisfy the personalized needs of clients, socialized information filteringemerges accordingly. Thus, personalized recommendation, which is based oncollaborative filtering recommendation technology, enjoys tremendous developmentin theory and application.Group recommendation, a new research spot in personalized recommendationfield, focuses on the group rather than the individual. The difficult part of conductingresearch on group recommendation lies in the complexity and heterogeneity of human.Usually group recommendation system cannot balance the preference of eachindividual well, thus its overall satisfaction degree is rather low. To solve the problem,one of the key steps is to negotiate with group members on the potentialrecommendation items. The present thesis applies case-based reasoning method andnegotiation theory to group recommendation system, simulates negotiation of therecommendation issue in the group, and proposes algorithm based on the idea.The thesis first elaborates on the theoretical basis and research status of grouprecommendation, personalized recommendation system, negotiation theory, andcase-based reasoning method. The existing judgment of group members acts as theknowledge base. Starts from the perspective of group members, the thesis first studiesthe recommendation issue of two-member group. After setting certain rounds ofnegotiation and degree of patience, the research stimulates the process of negotiationamong group member through Agents, and recommendation is made after reaching aconsensus. With respect to the recommendation issue of multi-member group, thethesis put forth that sub-group shall be formed. After the in-sub-group negotiation, aninter-sub-group negotiation can be done, and final recommendation can be achievedaccording to overall agreement. The feedback from clients can be utilized to timelyupdate the knowledge base. The last step is to evaluate the experiment by adoptingMovieLens database. The research results show that at different group scales thealgorithm of this thesis excels comparing algorithm in recommendation quality.