User's Coupon Usage Behaviour Prediction In E-commerce Environments

In the e-commerce environment,merchants usually increase their profit by issuing coupons to the users.A coupon issued by a merchant is returned to the merchant through the user's purchase behavior,which is called coupon verification,also known as redemption.There is no doubt that the verification rate of coupons is an important indicator to evaluate the success of a merchant's coupon marketing behavior.If the merchant does not develop a suitable coupon strategy,or randomly issues coupons to the customers,they may not take effects and thus waste the budget.It is very important for merchants to issue coupons to customers who are more likely to have consumer behavior.In this paper,we study the merchants' coupons behavior in the e-commerce environment.Specifically,the main works of this paper are follows:1)We studies the user's coupon behavior,we use the coupon behavior data of the customer in the real scene,and understand the data from the three objects: user,coupon and merchant.We designs a wide range of features,a total of 71 features were designed in the feature engineering of our work.The effectiveness of each group feature was verified by our experiments.The features designed by this paper can be a significant reference for the coupon behavior prediction task.2)We applies a variety of machine learning methods and builds a ensemble model system for the coupon behavior prediction task.We measure the effects of each model by using model pre-training,model evaluation.Then,we select the appropriate machine learning model,and construct a highly robust ensemble system for the coupon behavior prediction.3)We introduces a new type of deep learning network,"Capsule Network".The features designed by our feature engineering are used as multiple capsules,and two capsule network structure for predicting coupon behavior is proposed in our work.In this paper,the proposed two capsule network structure is compared with multi-layer perceptron,convolutional neural network and recurren neural network.The experimental results show that the proposed capsule network structures in this paper has strong superiority.In particular,this paper is the first time to apply the capsule network on the e-commerce recommendation system scenes...