Collaborative Filtering Recommendation Algorithm Based On User Feedback And Its Timeliness Improvement

The problem of efficient acquisition of personalized information is increasingly prominent in the era of information explosion,which has attracted extensive attention from all walks of life.Preference information and timeliness is different from person to person,including implicit feedback of user behavior can more accurate feedback from the user's preferences,and present a certain correlation,so the user's different interest preferences and targeted training model was constructed,the prediction information of interest to users,for users to obtain personal preference information is realistic.Collaborative filtering recommendation is one of the most widely used and studied recommendation algorithms in today's society.Therefore,the goal of this paper is to build a relatively excellent collaborative filtering recommendation algorithm model based on the adaptive combination of user interest and scoring preference differences,and study how to accurately recommend personalized preference information for users.Although the traditional recommendation based on content and association rules can improve the accuracy of prediction recommendation to a certain extent,due to its own limitations,the above method is still not superior in terms of prediction accuracy when dealing with sparse data and less display feedback from users.When processing sparse data,collaborative filtering recommendation algorithm can excavate users' implicit feedback and association more deeply,so as to achieve more accurate prediction effect.But most of the current collaborative filtering recommendation algorithm applies only to the user a total score a feedback to the user,such as portraits and associated modeling,ignoring the user of the total score and user behavior timeliness implicit feedback information,such as user's score is in modeling,did not take into account the user selection of reference frame for modeling the effect of graded habits,The limited information feedback of users is not explored and utilized to the maximum so as to provide users with more accurate recommendation and prediction.Therefore,the current research on collaborative filtering recommendation algorithm still has the following shortcomings :(1)it fails to fully mine the implicit feedback information of users by not fully considering the non-common scoring items of users,which cannot alleviate the data sparsity;(2)The timeliness of user behavior is not fully considered.When modeling user interest,the drift of user interest and attenuation of project popularity with the passage of time should be taken into account to improve the accuracy of recommendation;(3)In the modeling of user interest,the different scoring preferences and scales of different users should be taken into account.Compared with the standard scoring interval provided by the authorities,personalized modeling with higher reliability should be conducted based on users' personal scoring records.According to the above research situation,according to the characteristics of implicit feedback of user behavior,in the calculation of user similarity,in addition to considering the user's common score items,combined with all the user's item scores,the user's non-common score items were introduced into similarity calculation,alleviating the impact of data sparsity on prediction accuracy.Time factor is also integrated to alleviate the influence of user interest drift and project popularity attenuation on prediction accuracy.Combined with the coefficient of variation of user ratings,the behavioral preference differences of user ratings are modeled more accurately.On this basis,the improved user similarity and the difference of user rating behavior preference are adaptively integrated into the final user similarity,and the prediction results are generated according to it.The main contributions of this paper are as follows:(1)In this paper,an improved PCC similarity measurement(ITPCR)method was proposed,which not only considered the context information of user item score,but also incorporated the time factor to alleviate the effects of data sparsity,user interest drift and item popularity attenuation.(2)A method based on the coefficient of variation is proposed to model the difference in user rating preference(URP),which uses the user's historical rating behavior as a dimension to model user rating preference and improve the credibility of the model.The improved ITPCR similarity and user rating preference difference URP were combined into ITPCR global similarity by weighted adaptive combination,and more accurate prediction results were generated based on this;(3)RD and TD are proposed to improve the PCC correlation coefficient by combining user information with project information.RD and TD are normalized exponentially by SIGMOD function,and the improved coefficient P is constructed,which is combined with the traditional PCC correlation coefficient into the global similarity of IMPCC,and the final prediction accuracy is improved accordingly.