Research On E-commerce Recommendation Based On Multiple Machine Learning Algorithms

As we all know,recommendation system plays an indispensable role in today’s society.Based on a large amount of data,it uses algorithms to analyze the data and select the content or product that users like.In this way,it not only improves the user experience,enhances user stickiness,but also brings greater profit space for businesses.E-commerce is a popular marketing mode at present.In order to study the recommendation system in the field of e-commerce,this paper selects the behavioral data of 20,000 users in one month,including browsing,bookmarking,adding shopping cart and purchasing,and converts the prediction of users’ purchasing behavior after one month into a binary problem.Firstly,according to the interaction between the data of the nearest day to the target date and the historical data,the data is divided into the weekly dimension data,and the data is divided into the training set,the verification set and the test set.Secondly,four types of features were extracted from the source data,including counting,sorting,conversion rate and time difference.There were 107 features in total.The data of each group were labeled according to the purchase of data on the last day,where 0 represented no purchase and 1 represented purchase.Thirdly,due to the serious imbalance between positive samples and negative samples,we use k-means to cluster negative samples.Finally,Logistic regression,neural network and GBDT algorithm were used to train the training data set in the way of searching the optimal parameters manually,and the proportion of positive samples and negative samples of the training data was adjusted at the same time.From the experimental results,GBDT prediction score is the highest,0.119,Logistic regression and neural network prediction effect is not much different.