De-Biasing Rating Propensityalgorithmin Group Recommendation

Abstract

In recent years, group recommendation systems have gradually attracted attentionwith the increasingphenomenon of people's group activities.Nonetheless, most research focuses on optimizing machine learning models to fit user behavior databetter.However, user behavior data is observational rather than experimental. Due to the different psychological benchmarks of user ratings, the training data evaluated by the algorithm cannot fully represent the real preferences of the target group. A De-Biasing Rating Propensity Algorithmin group recommendation is proposed. The proposed algorithmidentifies user groups with similar behavior preferences through the Predict & AHC algorithm based on cosine similarity, and calculates user bias information by groupand user preference tendency for different user groups. The De-Biasing Proportionon different items is used to build a rating bias consistency model, which effectively adjusts the user's predicted rating.The experimental results show that the algorithm can significantly improve the quality and fairness of recommendation.