Research On Group Recommendation Algorithm Based On User Behavior Bias

With the rapid development of Internet technology,and the exponential growth of network services.It is difficult for people to obtain their own requirements from the accurately massive candidate services.Recommendation systems have become an important information filtering technology to alleviate information overload and help users to select out services they are interested in conveniently and quickly.However,the phenomenon of people moving in groups is increasing in real life.Therefore,the group recommendation systems has gradually attracted attention,and there are various potential biases in the data of the interaction behavior between user groups and recommendation services.Most of this interactive data is observational rather than experimental.Due to the influence of these potential biases,group member preferences are skewed,which leads to poor recommendation performance.Therefore,this thesis studies research work in analyzing the bias problem of user interaction behavior in group recommendation systems.The main work is as follows:(1)This thesis proposes a group recommendation algorithm based on the elimination of rating propensity bias.To address the bias inherent in group user interaction data and the problem of user interest propensity,different groups of users use group preference propensity and user preference propensity to represent user bias information respectively.An unbiased group recommendation model is constructed through the dynamic de-biasing process of rating data.The experimental results show that the algorithm has certain advantages in mitigating data bias,improving recommendation quality and recommendation fairness compared with the traditional group recommendation algorithm improvement strategy.(2)This thesis proposes a group recommendation algorithm based on eliminating user conformity bias.Aiming at the problem of popularity bias in group recommendation system,this thesis starts from the user level and item level.Fusing user interest propensity and item popularity bias to represent user herding effects,and calibrating group recommendation results based on the debias factor.Identify beneficial popularity bias and construct a group preference model based on the de-biasing factor.The experimental results show that the algorithm identifies beneficial prevalence deviations and improves the recommendation accuracy and coverage.(3)This thesis designs and implements a group recommendation prototype system based on the above proposed algorithm.Follow the general steps in the requirements analysis to build the prototype system such as modular functional realization.The theoretical value and interpretability of the algorithm proposed in this thesis are verified in realistic scenarios.