| In recent years,e-commerce technology has developed rapidly,and more and more users are able to "zero distance" to buy items of interest.The data of major e-commerce platforms has also increased at an exponential rate,including user reviews and rating data,item soft text data,and item tag data.In the complicated and massive data,how to quickly and efficiently recommend suitable items to users has become an urgent problem to be solved by current e-commerce platforms.In this context,personalized recommendation systems appear.The volume of tea products on e-commerce platforms continues to increase.Because it is difficult to obtain effective information to characterize the individual characteristics of users,natural language processing(NLP)and deep learning(DL)methods are used to carry out users.The characterization of characteristics,and thus better implementation of tea product recommendation,has certain significance for the development of tea product e-commerce.Based on the analysis of the relevant theories of personalized recommendation,this paper proposes a personalized recommendation model of tea products based on BERT-BLSTM in view of the shortcomings of tea product recommendation in existing e-commerce platforms.The specific work is as follows:(1)According to the previous feature representation method,it is unable to make full use of the context information in the tea product review text and cannot solve the problem of word ambiguity.Based on the tea product review text,the BERT model is introduced for parallel semantic processing of word granularity,so that after embedding'S vectors can better reflect the relationship between words and words to represent user feature preferences.(2)Aiming at the problem of the disappearance of the gradient of the traditional recurrent neural network and the lack of obtaining reverse semantic information of the unidirectional neural network,a bidirectional long short term memory network(BLSTM)is used to train tea products from both positive and negative sequence directions Comment text to obtain the word order representation of the comment text.According to the text features obtained by training,multi-layer perceptron(Multi-Layer Perception,MLP)is used for score prediction.By making full use of the semantic information in the comment text,the performance of scoring prediction is improved.(3)On this basis,a personalized recommendation model for tea products based on BERT-BLSTM is proposed.Experiments with real data sets collected on the Jingdong platform verify the effectiveness of the model proposed in this paper. |
