A Recommendation Framework Based on Predicting User Dynamic Preferences
Time data is easily accessible and present in various applications.However,most recommenda-tion methods fail to consider the dynamic changes in user preferences,leading to outdated and impersonalized recommendations,resulting in challenges such as decreased user experience and information overload.To address these issues,this study proposes a recommendation framework based on Predicting User Dynamic Preferences(PUDP)using historical purchase sequence data.The framework consists of three stages:first,user features are obtained using matrix factorization and purchase time series data;next,the Kalman filter is employed to predict user preference vectors from these features;finally,two recommendation sequence calculation methods are proposed based on the predicted preference vectors to generate recommendation lists for users.Experimental results demonstrate that,on a real-world Last.fm dataset,this method outperforms competing methods such as the first-order Markov model in recommendation prediction performance.
matrix factorizationtime-aware recommendationkalman filterrecommendation system
万方数据
维普
