Research On Methods Of Basket Recommendation Based On Deep Learning

With the popularity of smart phones,e-commerce has developed rapidly.Shopping basket recommendation is an extremely important task in the field of e-commerce.The researchers proposed the use of association rules analysis,collaborative filtering,Markov chain,recurrent neural network and other methods for shopping basket recommendation.However,these methods ignore some shopping basket data features,which leads to the need to improve the accuracy of recommendation results on the one hand,and the coverage and novelty of recommendation results on the other.This paper designed three shopping basket recommendation algorithms based on the characteristics of shopping basket data.The main research contents and innovations are as follows:1.Follow the commonly used TOP-K recommendation task of shopping basket recommendation.In view of the characteristics of shopping basket data,this article extends a self-attention mechanism.On the one hand,this mechanism can fully consider the problem of matching between different products in the same shopping basket On the other hand,this mechanism is used to integrate the shopping basket product attribute information into it.Aiming at the characteristics of shopping basket data sequence,this paper uses recurrent neural network and one-way self-attention to capture the long-term and short-term characteristics of the sequence.On the three real data sets,the method proposed in this paper is superior to the traditional method in accuracy.2.The accuracy of the recommendation results in the TOK-K recommendation task has improved,but the result coverage and novelty are low.This article proposes to use the generated method to recommend shopping baskets.This article uses recursive recommendation tasks,recommending one product at a time and recursively recommending K times.This article uses the classic Encoder-Decoder architecture of the generative model.In order to solve the error transmission and repeated prediction problems in the Decoder,this article extends two solutions.On the two real data sets,the method proposed in this paper has greatly improved the coverage and novelty of the recommendation results when the accuracy of the recommendation results has not decreased significantly.3.In the first two ways,the network-optimized KL divergence has an asymmetry problem.This article proposes to use adversarial generation as a shopping basket recommendation.Using the generator-discriminator architecture,the generator predicts the items in the next shopping basket,and the discriminator determines whether the predicted items are actually purchased.This paper extends the optimization goal of the adversarial generative model.Regarding the problem that the gradient between the discriminator and the generator cannot be transferred,the strategy gradient of reinforcement learning is used.On the two real data sets,the method proposed in this paper has further improved accuracy.In this paper,three shopping basket recommendation algorithms are designed according to the shopping basket data features,and experiments are carried out on real data sets.The experimental results show that the shopping basket recommendation algorithm designed in this paper is more accurate and of higher coverage than the traditional recommendation algorithms.