Commodity Potential Customers Mining And Accurate Recommendation Based On User Behavior

Under the background of the rapid development of the Internet industry,the expansion of the scale of e-commerce makes more and more manufacturers join the e-commerce industry in line with the call of The Times,which virtually increases the difficulty for users to choose commodities reasonably.E-commerce platforms generally use recommendation algorithms to complete recommendation services for users or merchants,but traditional recommendation algorithms are not effective in solving the problems of cold start of recommendation system and user interest transfer.Using user behavior data to evaluate user interest is an effective method to solve this kind of problem.However,most algorithms only involve the browsing data of users,so they cannot accurately grasp the real-time interest of users,and their interest measurement is not accurate enough,resulting in low accuracy of recommendation system.Based on the traditional personalized recommendation algorithm,this paper aims at the problem that the accuracy of measuring the change of users’ interest is not high,and USES the analysis of users’ various behavior data to complete the prediction of potential customers.Mainly done the following work:Firstly,the research status of personalized recommendation algorithm was analyzed.Aiming at the low accuracy of recommendation caused by cold startup and user interest transfer,the main data indicators reflecting user interest were obtained.According to contain commodity number,attributes,the brand’s product data,and contain the user number,user interaction,commodity code,interaction and buy index of information user behavior data,design includes commodity attribute,category and activity of three data index model and user model,and further using the principal component analysis to the rationality of the design of algorithm is verified.Then,considering the different Settings of user interest indicators of user behavior data,the accuracy of measuring user interest is improved.On the basis of the preliminary similarity calculation of the commodity model and the user model based on the commodity model,the user interest value index is designed according to the different tendencies reflected by the user interest,such as adding purchase,deleting purchase,etc.,and the final user interest value is obtained by adding with the preliminary similarity.Finally,thecoefficients in the algorithm design are assigned by experiments and compared with the traditional personalized recommendation algorithm.Through the experiment,the threshold value?(28)6 is set for the user’s commodity interest value,and the user whose interest value of the commodity exceeds this threshold is taken as the potential customer of the commodity.Compared with the traditional recommendation algorithm and machine learning recommendation algorithm,this recommendation algorithm based on user behavior data concludes that the recommendation accuracy rate is 92.80%,recall rate is 39.80% and F value is 55.70%.Based on the analysis of indicators,this paper expands the types of user behavior data to more accurately assess user interest and achieve better recommendation effect.It provides a theoretical basis for the platform to better stabilize customers,expand the market and provide better service for users by improving the satisfaction of users’ demands.