Research On Sentiment Classification Based On Emotional Dictionary And Deep Learning Technology

With the rapid development of E-business and mobile Internet,customers can buy almost anything no matter where it come from.However,millions of products on electric business platform also bring the "products overload" problem to customers.Even after strict classification and directed search,the customers still cannot quickly obtain the information about the products what they want.Therefore,a "salesman" is urgently needed to provide recommendation service to customers.Electricity recommendation system can generate personalized result from mass products by mining the binary relation between customers and products.This system can effectively improve customers shopping experience and enhance the acceptance and adhesiveness of customers to E-business platform.As one of the important application fields of recommendation system,e-commerce is a complex system engineering which needs to be considered from the big data technology,system architecture and arithmetic.The major work in the paper and innovations are as follows.(1)Analysis of the current big data technology of the distributed storage and resource management,On the basis of the current Hadoop version has master node single-point trouble in the HDFS,put forward the high availability of HDFS improvements,and introduces in detail how to elastic computing resource management.(2)Research the current mainstream recommendation system,Analysis the architecture and trigger process about recommendation system and points out that the traditional recommendation system architecture and trigger process has many defects,puts forward improved recommendation system architecture and trigger process to improve the performance of recommendation service.(3)According to the State negative feedback data is ignored in traditional matrix decomposition model,which leads to the recommendation accuracy cannot be improved,the matrix decomposition model with negative feedback data set is a solution.The model by using the AB Testing method to measure the implicit feedback ratings to build implicit feedback rate score matrix,the data set to join the negative feedback data,through the use of new loss function to make the best fitting model and the data set.The online comparison results show that the model effectively improves the clicks and transformation.(4)Current E-business commerce recommendation systems can provide personalized recommendation service for buyers,but does not take into account the goods quality,for this circumstance is proposed,in this paper a label based weighted association rules mining algorithm,provides the high quality of the personalized recommendation for the buyer.Through the off-line experimental analysis,the proposed algorithm is better than the traditional mining algorithm of association rules.