Research On Personalized Collaborative Filtering Recommendation Algorithm Based On Users' Preference

With the advancement and development of information technology,the problem of "information overload" has become increasingly prominent.This problem leads to a significant increase in the time and labor costs for people to obtain the required information.The most effective way to solve this problem is personalized recommendation.Collaborative filtering is one of the most widely used recommendation algorithms.After a lot of research and improvement,the algorithm still has the following problems that affect the accuracy of recommendation results:(1)sparse rating data;(2)the neighbor selection is unreasonable due to different users' rating standards;(3)it is difficult to capture the changes in users' preferences and interests.In response to the above problems,the main research work of this thesis is as follows:1.To solve the problem of sparse scoring data,the commonly used solution currently is filling the sparse matrix.However,these filling solutions do not consider the users' preferences and the items' differences fully.This thesis improved it,and proposed an improved filling algorithm based on users' preferences.The algorithm improved the problem from two aspects.Firstly,it calculated the users' preference weight and average value of different items' attributes based on the users' ratings,and then used the above results to calculate the padding values and fill the matrix.Secondly,according to the different target user,it improved the calculation of items' similarity to obtain the more reasonable neighbors.2.It can be seen from the analysis of the problem of rating differences,this problem mainly affected the reliability of the similarity calculation results,resulting in unsatisfactory recommendation results.This thesis improved it and proposed an improved recommendation algorithm based on the rating differences.The algorithm firstly used the modified cosine similarity calculation to take the user's average rating as the scoring criterion.Then,it introduced users' impact factor and items' impact factor into the similarity calculation to correct the influence of different users' rated items and different items' quality on the similarity calculation.3.By analyzing the problem of the changes in users' interest,it can be seen that the traditional algorithm does not distinguish the different rating time of different items,which leads to the problem of low timeliness of recommendation.This thesis improved it and used the time factor to characterize the changes in users' interest.According to the impact of time on the users and the items,the impact between different items,and the impact between different users,three time decay functions were constructed.And then,by weighting the improved hybrid recommendation algorithm based on the users' preferences and rating differences,an improved algorithm which considers the time factor is obtained.Finally,the Movie Lens dataset is used to verify the improved algorithm proposed in this thesis.Experimental results showed that the improved algorithm proposed in this thesis can effectively improve the accuracy of recommendation.Compared with the traditional algorithm,the recommendation accuracy of the UPRDB algorithm based on the combination of users' preferences and rating differences has improved by an average of 7.87%.On the basis of the UPRDB algorithm,the recommendation accuracy of the UPRDB?T algorithm which proposed in consideration of changes in users' interest has improved by an average of 2.33%.