Design And Implementation Of Commodity Recommendation System Based On User Portrait

In recent years,with the vigorous development of e-commerce,the structure of e-commerce website is more complex,the scale is larger,and the types of goods are also increasing.How to recommend suitable goods for users in a large number of goods has become an urgent problem for service providers.After collecting users' static data(users' registration information,etc.)and dynamic data(users' browsing records,shopping lists,etc.),the commodity recommendation system based on user portrait uses statistical method to determine the information portrait for static data,and uses textrank algorithm to generate behavior tags for dynamic data,The TF-IDF algorithm is used to calculate the weight of the behavior tag to generate the behavior portrait,and the user portrait is determined by the information portrait and the behavior portrait.When TF-IDF algorithm is used to calculate the weight of behavior tags,only the relationship between users and tags is considered,and the influence of the correlation between tags on the weight of tags is not considered.At the same time,the particularity of e-commerce scenarios is not considered.Therefore,this paper proposes a user tag weight calculation method in the e-commerce scenario.Based on the TF-IDF algorithm,the correlation coefficient matrix is used to reduce the influence of correlation on the weight calculation.Considering the different influence of user behavior types on the weight in the e-commerce scenario and the interest will decline with time,the user tag weight is further processed in the scenario,Get more realistic and reliable user portrait.Finally,the user's interest in the product is calculated quantitatively by the user's portrait,and the recommendation list is obtained after sorting.The system is designed and implemented around the user's portrait and product recommendation.At present,the ndcg value of the optimized tag weight calculation method is increased by 5 percentage points,which can reach 0.87.In the future,we can increase the sentiment analysis of the comment information to make the user profile more realistic and representative,so as to get a more satisfactory recommendation list.The establishment of a real and representative user tag model can explore the unpopular content services,tap the "long tail" benefits,and improve user loyalty.