Research On Personalized Recommendation Core Technology Based On WEB User Behavior

With the development of social informationization,online shopping has become an indispensable part of people's life because of its advantages of complete commodity types,transparent price and convenient distribution.In this era of Internet,which focuses on service experience,the quality of user experience largely determines the rise and fall of a platform.At present,the optimization of platform services has become an important means for e-commerce platforms to drain customers and solidify customers.Among many user services,personalized recommendation,as an important means of commodity recommendation,has been criticized by users for a long time.The core reason is that the accuracy of recommendation is too low and traditional personalized recommendation is too low.The system mainly has the following problems: inadequate recommendation accuracy,waste of system resources,low accuracy of data,single recommendation model,etc.In this paper,by capturing the overall behavior of users during their visits and analyzing and extracting the behavior,we can get a more accurate user behavior feature model,and then improve the accuracy of the recommendation results,because each visit of users will have a large number of behaviors recorded and analyzed,which greatly reduces the simplicity of the recommendation model.Problem.In this paper,the analysis of behavior characteristics and model extraction are independent by means of distributed method,and the parallel processing of multi-threaded simulation is used to further improve the operation speed and reduce the waste of system resources.The traditional data processing work is reduced to zero.The data are processed in real time using distributed and multi-threaded simulation parallel processing method.The behavior analysis and feature extraction module is independent,which effectively improves the stability,security and scalability of the system as a whole.At present,many small and medium-sized e-commerce platforms do not have high concurrent access under normal conditions,and the performance of servers is very considerable.Therefore,instead of choosing the traditional large data parallel framework,this paper adopts a lighter distributed multi-threaded simulation concurrent processing technology.The main work is as follows:1.Research and implementation of user access behavior capture method.2.Research and implementation of behavior data transformation and feature model building methods.3.Research and implementation of the output of behavioral feature model denoising.Through testing,the proposed personalized recommendation method improves the recommendation accuracy to a certain extent,and reduces the dependence on historical data sets,the amount of data calculation,and the cost of functional implementation to a certain extent.