Research And Application Of User Profile Technology In E-commerce System

This paper studies the user profile technology in e-commerce,and focuses on the short text information classification and label weight calculation in user profiles.The Naive Bayesian text classification algorithm based on TF-IDF-DL and a hybrid algorithm combining TF-IDF and correlation coefficient matrix are designed.The two algorithms supplement the non-quantitative information classification and the weight inconsistency of the label weight calculation in the user profile technology.In the technical research of user profiles,the research on non-quantitative information is relatively rare,but in most cases,non-quantitative information can more strongly represent the user's preferences,such as user review information.The user comment information is mostly short text information,so the Naive Bayesian text classification algorithm is often used in classifying the user comment information,but the weight of all feature words is consistent in the process of feature calculation and feature classification,which will cause inaccurate classification.Aiming at this problem,this paper designs the Naive Bayesian improvement algorithm based on TF-IDF-DL from the frequency of feature words,the relationship between the position and category of feature words.The algorithm decentralizes the word frequency and introduce the influence factor of the position of the feature word in the calculation to improve the accuracy of text classification.Most of the previous studies have considered that the weight of the label is consistent,which may result in a problem of low accuracy for users to provide personalized recommendation services and a large amount of data loading.In order to solve this problem,this paper designs a hybrid algorithm combining TF-IDF and correlation coefficient matrix.The algorithm can reflect the influence of the label on the user from the number of labels.At the same time,it can reflect the correlation between labels and labels to a certain extent.The algorithm can achieve the effect of relative dimensionality reduction in the image technology,and can truly and accurately obtain the influence of the label on the user.Finally,this paper uses the data scene of a B2C e-commerce platform and the above algorithm.At the same time,the relevant experimental system was designed and developed,and the profile scheme was applied.The experimental system not only realizes the function of classifying the text comment information to the user,the function of calculating the reasonable weight of the label according to the user label,the visual display of the generated user profile,but also realizes the function of providing personalized recommendation for the user according to the profile,and the rationality of the function of the profile is verified by the recommendation result.