Research On Personalized E-commerce Product Recommendation Based On Deep Neural Network

With the rapid growth of content for users and product,it is increasingly difficult for users to obtain products of interest by means of manual search or view one by one.In order to improve the user consumption experience and efficiency,the Internet platform uses the recommendation system to help users purchase.Using the recommendation algorithm to dig out the products that the user most likely likes to push.These algorithms have their own limitations in dealing with data sparsity problems and cold start problems,the use of deep neural networks in recommended areas can effectively solve the above problems.In this dissertation,the model based on latent features is effectively combined with the deep neural network,and proposed two hybrid recommendation algorithms which can simultaneously mine features from product content information and historical interaction information,they can be applied to commodity recommendation in different life cycles.Finally,a recommendation framework based on deep neural network is designed for the different characteristics of product,and suitable algorithms are matched for different commodities to improve the accuracy of recommendation.The main work of this dissertation includes:(1)The development history and research status of traditional recommendation algorithm combined with deep learning technology are analyzed in detail,and pointing out the shortcomings in current research.The reasons for these shortcomings and the solution ideas are given.This dissertation introduces the relevant theoretical foundations involved,and also talks about the application,theoretical background and practical application of collaborative filtering algorithm,content-based recommendation and deep neural network.(2)The new product lacks historical interaction data,by introducing image data and using convolutional neural network to extract commodity visual content features as auxiliary information of recommendation system,combined with hidden potential model to mine potential features of products from scoring data,the two types of features complement each other.The convolutional neural network and implicit semantic model are integrated and a novel hybrid recommendation algorithm is proposed,through the comparison experiment under the real data set,the algorithm is obviously superior to other models,when the data is scarce,the model shows excellent stability.(3)In order to improve the recommendation effect of mature products,the review text data is added on the basis of the score data,and the long-short-term memory neural network based on the attention mechanism is used to extract the text characteristics from the product's comment text to characterize the potential characteristics of the product.Making up for the semantic gap between text and recommendation systems.Then,it combines with the features from the latent semantic model,and restricts and complements each other.Finally,the hybrid recommendation algorithm is compared by real-world real data sets.The experimental results show that the algorithm is superior to other advanced algorithms and improves the accuracy of scoring prediction.(4)The proposed deep neural network-based recommendation algorithm is integrated into a recommendation framework,and the products are subdivided according to different life cycles of the products,and different recommendation algorithms are used for training and prediction,which effectively improves the accuracy.In addition,the recommendation framework makes full use of the data resources of today's Internet era,adds unstructured data with a large amount of information,and enriches the data input of the recommendation system.The recommended performance of the recommendation framework based on deep neural network is evaluated by comparative experiments.The experimental results show that the recommended performance of the proposed framework based on deep neural network is better than the existing recommendation framework.