Research On Fine-grained Sentiment Analysis Method Of Cross-domain E-commerce Text Reviews

With the continuous maturity of e-commerce technology and the continuous expansion of online shopping scale,more and more consumers use online shopping to replace offline shopping in physical stores.In the environment of online shopping,consumers cannot completely trust the product information provided by the merchant,and then they will take the e-commerce text reviews as the standard to judge the quality of the product.E-commerce text reviews contain a large number of subjective emotions,which are divided into positive,negative and neutral according to emotional polarity.At the same time,the subjective sentiment contained in e-commerce text reviews will involve multiple aspects.In order to obtain the emotional tendency of a certain aspect of the product,fine-grained sentiment analysis technology is needed.Therefore,based on the research at home and abroad,this article takes e-commerce text reviews as the research object,analyzes the emotional polarity of different aspects,and solves the problem of field data scarcity,and finally realizes the finegrained sentiment analysis of cross-domain e-commerce text reviews.The research content mainly includes the following aspects:(1)Aiming at the problem of fine-grained sentiment analysis of cross-domain ecommerce text reviews,the goal of this task is to classify different aspects of e-commerce text reviews to obtain the sentiment polarity of different aspects.In this article,ERNIE-Gram pretraining model is used to extract the features of e-commerce text reviews,and it is input into the improved DPCNN neural network to construct the ERNIE Gram-DPCNN joint model,and then the pre-training model is fine-tuned to realize the fine-grained sentiment analysis task of cross-domain e-commerce text reviews.Finally,through experimental comparison,it is verified that the precision of ERNIE Gram-DPCNN model is higher than other models.(2)Aiming at the problem of data scarcity of cross-domain e-commerce text reviews,a fine-grained sentiment classification model of cross-domain e-commerce text reviews SKEP Gram-CDNN based on generative adversarial network is proposed.The model takes generative adversarial network as the framework.The emotional knowledge obtained from the pre-trained SKEP model and the e-commerce text features extracted from the Ernne-GramDPCNN model are fused as the generator of the generative adversarial network,and according to the superior performance of the capsule network in the cross-domain transfer task,it is used as the discriminator of the generative adversarial network and applied to the cross-domain finegrained sentiment classification task.By transferring the source domain knowledge to the target domain,it effectively solves the problem of data scarcity in some e-commerce fields,meets the diverse needs of practical applications,and promotes the cross integration of disciplines.Finally,the effectiveness of SKEP_Gram-CDNN model is proved by comparative experiments.