Research On Personalized Recommendation Method Based On Dynamic Changes Of User Preferences

Increasing amounts of information are coming along with the era of big data,making it difficult for users to find information that they are interested in or are useful for themselves with limited time and energy.The emergence and development of the recommendation system provide great help for people to solve the difficulty of choosing redundant information.It can effectively mine users' interests and preferences according to their past behavior information,and then provide personalized recommendation services.Generally speaking,users' preferences will change with time.Therefore,how to better mine users' preferences and track the dynamic changes of users' preferences over time positively affects user personalized recommendations' effectiveness.Based on these,this article takes the 100,000 ratings of 1,682 movies from 943 users in the sub-dataset ml-100 k of the public data set Movie Lens as the research object.The time factor is taken into account in the traditional collaborative filtering algorithm,improving personalized recommendation accuracy by portraying dynamic changes of users' preferences.Specifically,this article starts with the analysis of the rating matrix of users in the data set.Firstly,because of the high sparseness of the user-movie scoring matrix,the mode of user ratings and the Singular Value Decomposition(SVD)method are used to fill and correct blank items in the scoring matrix to obtain a highdimensional dense scoring matrix;secondly,we use the Principal Component Analysis(PCA)to reduce the above-mentioned high-dimensional dense matrix's dimensionality and perform the Kmeans clustering analysis on the reduced-dimensional matrix.According to the clustering results,users in the high-dimensional dense matrix are grouped and identified,which are used as the basis for subsequent data analysis;what's more,the users' access frequency and rating weight of a certain type of movie are used to reflect users' preferences,and the time is considered according to the user's rating time of the movie,combining the users' access frequency and rating weight of a certain type of movie,the rating time,and other factors to form a comprehensive indicator,which is used to describe dynamic changes of users' preferences;finally,we integrate this indicator representing dynamic changes of users' preferences into the traditional collaborative filtering method,weighting predicted scores through this indicator.Thus,we obtain a predicted score closer to users' actual preferences,forming a personality recommendation method based on dynamic changes of users' preferences.And the corresponding evaluation criteria for this recommended method are given.On the basis of previous,this paper conducts an empirical analysis with the selected data set.The empirical results show that this paper's personalized recommendation method has a certain reduction in recommendation error compared with the traditional collaborative filtering method.Among them,the Mean Absolute Error(MAE)and Root Mean Square Error(RMSE)were reduced by 5.41% and 8.06%,respectively,indicating that the processing method of sparse matrix and the characterization of dynamic changes in users' preferences in this article are essential for the final personalized recommendation.The research method of this article has certain reference significance for the research of relevant recommendation problems.