Research On Deep Collaborative Recommendation Of Tea Products Based On Attention Mechanism

With the rapid development of agricultural products e-commerce,tea products,as a kind of featured agricultural products,are trying to open up e-commerce channels,from offline to online,through the network to thousands of families.At present,the recommendation system is more and more widely applied and researched in the e-commerce of movies,books and so on,and it will also become the inevitable trend of the development of e-commerce of tea products in the future.The current tea product recommendation system,which cannot identify the difference of the user's concerns,and the identification of the user's concerns is of great significance to further explore the user's preferences and improve the recommendation accuracy.In deep learning,attention mechanism can improve the efficiency of paying attention to key features by re-assigning the weight of feature vectors.Therefore,it is the focus of this paper to improve the ability of collaborative filtering method to pay attention to key features in sparse data by taking advantage of the characteristics of attention mechanism.This paper takes the e-commerce of tea products as the research object and extracts user preferences and tea product features based on the review information of tea products.The main research contents are as follows:(1)For the current tea product recommendation system,little consideration is given to the influence of the interaction information between users and tea products on the focus characteristics.This paper proposes an improved model of interactive Pair-Attention.When calculating the attention weight of users' features,the influence of tea product features is added to obtain users' concerns.When the attention weight of tea product features is updated,user features are used as auxiliary information to explore the influence of interactive information on the acquisition of tea product concerns.(2)A deep collaborative recommendation model for tea products based on attention mechanism was proposed.The model uses the word vector matrix formed by the comment text as input.The Text vector matrix feature extraction is carried out by Text-CNN network with convolution kernel of different sizes.At the same time,combining with Pair-Attention,the feature vectors of users and tea products are weighted to play a role of paying attention to key features.Finally,the user's score prediction and recommendation task is completed by the multi-layer perceptron.(3)The collected real tea product data set was used for experimental simulation toverify the effectiveness of the interactive attention mechanism Pair-Attention proposed in this paper,and the influence of different parameter Settings on the prediction accuracy of tea product score was explored.Comparing the proposed algorithm with the traditional matrix factorization algorithm,the multi-layer perceptron algorithm and the extended singular value decomposition algorithm,the experimental results show that the proposed method of tea product recommendation with interactive attention mechanism is superior to the traditional method in performance.