Research And Implementation Of Performing Arts Trading System Based On Personalized Recommendation

With the continuous development of the Internet and computer technology,e-commerce has greatly changed people's lifestyle.Faced with the colorful items on the major e-commerce platforms,it becomes a problem that how to effectively "match" the items to those who need.An indispensable tool for solving the problem above is recommendation system.The huge purchasing power and demands give the recommended system industrial value.However,the traditional recommendation technology is difficult to effectively solve many problems like data sparsity,cold start,real-time recommendation and so on.In addition,the engineering implementation of the recommendation system in the context of big data has always been a research difficulty and hotspot in the field.In order to solve the above problems,relying on the extensive application of deep neural networks in the field of intelligent recommendation,this thesis proposes a deep network model using Long Short-Term Memory(LSTM)neural network and Multi-Layer Perceptron(MLP),which mainly uses the description information of users and items combining with decompose user-item rating matrix to alleviate the sparseness problem of the user-item rating matrix and the cold start of the user and the item.Secondly,after getting the relevance between users and items,in order to make the recommendations more personalized,the Aspect and Sentiment Unification Model(ASUM)is used to mine the user's emotional theme topics and adjust the order of recommendation list.Finally,based on the proposed recommendation model,combined with big data technology design to realize the performing arts trading system.The main work of this thesis includes the following:(1)A feature fusion deep neural network(Feature Fusion Deep Neural Network,FFDNN)is proposed.The model first constructs and fills the user-item rating matrix,and uses the matrix decomposition technology to decompose the rating matrix,and then constructs the feature vector of user and item respectively by fusion the description information of the user and the item.The LSTM is used to deal with user feature vector and project feature vector to generate effective potential features,and to mine the potential relationship between users and projects,which effectively alleviates the sparse rating matrix and the problems of user cold start and item cold start.It effectively alleviates the problem of sparseness of the rating matrix and the cold start of the user and the cold start of the item.Using MLP to reduce the potential features generated by the LSTM and to improve the nonlinear modeling ability of the model as well as predicted the correlation between user and item according to the network output of the MLP,and the predicted association degree score is used as a basis for generating recommendation lists to complete the recommendation.The specific model design is given in the thesis and other models are compared on the public dataset.The results show that the proposed model has higher prediction accuracy.(2)The ASUM model is implemented and used to process user comments and extract the probability distribution of user emotional topics.Specifically,it constructs the ASUM corpus and conducts comment analysis,mines the user's likes and dislikes of each item theme,calculates the prediction result generated by the FFDNN model,adjusts the recommendation list`s order generated by the FFDNN model,and generates new recommendation results to the user.(3)Based on the proposed algorithm model,a performance trading system(hereinafter referred to “YanYiTong”)is designed and implemented.The system is an e-commerce platform for trading performance activities as commodities and which greatly shows the actual use value of the algorithm proposed.Facing with the continually growing unstructured data,processing data with traditional relational database is complicated and inefficient.This thesis designs an information storage mode based on secondary index in HBase database,and combines arbitrary fields according to actual needs to speed up the query.Spark Streaming,a big data real-time computing framework is applied to process data in order to ensure real-time performance of the recommendation system.