Research And Application Of Recommendation Algorithms Based On Preference And Nearest Neighbor Iteration

In the era of big data,Internet information showing an exponential growth,to bring people to a wide range of network life,but also to the user brings the corresponding problems.One of the most important problems is to get the information of interest in the mass information.The recommendation algorithm and the search engine are the effective methods to solve this problem.Recommendation algorithm based on the user's historical behavior records to predict demand to provide recommendations,relative to the search engine is more humane.Although the proposed algorithm has been widely used in various fields,but there are still some shortcomings.Such as cold boot problems,the proposed algorithm can not provide accurate recommendations for new users and new projects;recommended precision and diversity issues,due to data sparseness,etc.,resulting in lower recommendation accuracy.In this paper,we propose the corresponding improvement and innovation on the basis of collaborative filtering algorithm for the accuracy of recommendation algorithm.Based on the existing algorithm,the user preference model is reconstructed and the influence of user rating on the preference tag is introduced.In order to get a model and the user's actual preference,a new algorithm is proposed to solve the problem of user preference.And then weighted on the basis of the prediction score using the user collaborative filtering algorithm.The influence of user preference is enlarged by the prediction weighting,which avoids the influence of user preference caused by multiple calculation and trade-off.Neighbor iteration recommendation algorithm combined with the traditional two nearest neighbor selection method,first use TOP-K method to select the user's neighbor set,and then use the threshold filter to filter out the close neighbors,and then select the neighbors according to neighbors.On the basis of the second-level neighbor,the attenuation ratio and the attenuation weight are introduced to recalculate the similarity among the users to form a candidate neighbor set,and then iterate the neighbor selection and optimize the neighbor selection result.Combined with the above two algorithms to calculate the user predictive value,the forecasting tendency is used to modify the prediction score,which makes the score closer to the user's true score.Experimental results show that the proposed algorithm can improve the recommendation accuracy and alleviate the problem of scoring matrix.