Article Preview
Top1. Introduction
Similar to traditional e-commerce platforms, if community group buying companies want to improve their competitiveness, they need to build an excellent personalized recommendation system. The premise of recommendation is prediction. At present, in terms of prediction, major e-commerce platforms have started to use advanced technologies such as big data mining and recommendation algorithms to mine users' historical behaviors. With the rise of big data and artificial intelligence, many scholars have tried to use a combination of big data technology and machine learning algorithms to study the purchase behavior of traditional e-commerce users, aiming to improve the robustness and prediction accuracy of the model.
Du and Huang (2015) improved the decision tree algorithm and built a new user purchasing behavior prediction model, which achieved the global optimum from the local optimum. Li and Qi (2016) constructed a Pareto/NBD prediction model based on the user data on Dianping.com, and improved the prediction effect of the model on user purchase behavior by considering covariates. Zeng (2017) constructed a selection model based on the hesitant psychology of users when consuming, combining latent factors and behavior sequences, realized the prediction of consumer behavior, and verified the effectiveness of the algorithm. Zhang (2019) improved the Bagging ensemble learning method and proposed a subdivided ensemble learning method. The final trained subdivision ensemble learning model has better prediction effect than the existing machine learning methods. Duan (2020) used the K-means algorithm to improve the random undersampling data balance method, and on this basis built a long-short-term memory network algorithm, an extreme gradient boosting algorithm and a logistic regression algorithm as the base classifier. The boosting algorithm is a fusion model for user purchase behavior prediction of meta-classifiers. Xu (2018) proposed a two-layer fusion model algorithm based on the GBDT model according to the ensemble learning idea in machine learning to predict the repurchase behavior of users. Experiment results show that the fusion algorithm improves the prediction accuracy and robustness. sex. Mou et al. (2017) used the pricing framework to develop and test the buyer's repeated purchase intention model in cross-border e-commerce. The research shows that value, currency savings, convenience and product supply have the greatest impact on repeated purchase intention. Zhao et al. (2018) proposed a merged model of different classification models and a LightGBM fusion model with different parameter sets, and the experimental results show that the merged model can bring great performance improvement compared to the original model. Diamantaras et al. (2021) proposed a method for real-time prediction of e-commerce users' shopping intent using LSTM recurrent neural network. Cai et al. (2020) used the recurrent neural network algorithm to classify the user behavior sequence, obtained the user behavior propensity score and used the score as a new feature, and used the naive Bayes algorithm to optimize the recurrent neural network, and finally obtained the ratio of a single naive Bayesian model with higher prediction accuracy. Wang et al. (2018) proposed an improved algorithm that combines Logistic regression and XGBoost algorithm to predict user purchase behavior in e-commerce websites. The research shows that the logistic regression based on XGBoost method is feasible, and the composite method is better than two original methods when it is used alone.