Research And Implementation Of Cross-Domain Sentiment Classification Algorithm Based On Deep Learning

With the popularity and development of the Internet,more and more users are accustomed to commenting on the Internet.These texts contain various emotional tendencies,which mean great commercial value for both users and businesses.Because the amount of emotional data is large and disorganized,sentiment classification techniques are needed to quickly and efficiently mine emotional information.In addition,because the domains of comment data are more and more extensive,the review data in a domain may be insufficient,and the classifier trained in one domain cannot be directly used in another domain.Therefore,cross-domain sentiment classification is a worthwhile study.This paper mainly uses the deep learning method to classify cross-domain sentiment,and studies three problems and designs three deep learning models.In order to solve the problem that the deep learning model requires a large-scale corpus due to the large number of parameters,but the cost of constructing large-scale training data is expensive,this paper proposes a deep learning model based on multi-task learning.In the case of tagged data in multiple domains,the multi-tasking learning framework is used to comprehensively get domain-specific feature representations of multiple domains,and sample filtering and parameter transfer are used to obtain better classification results.In order to solve the problem of different representations of different domains in the cross-domain sentiment classification,this paper proposes a deep learning model based on attention mechanism and adversarial training.For the case where there is a large amount of tagged data in the source domain and the target domain has no tagged data,the attention mechanism is used to extract more accurate semantic features of the text,and the adversarial training is used to ensure that the obtained feature representations are shared between domains.Finally cross-domain sentiment classification is performed.In order to make full use of the untagged data of the target domain in cross-domain sentiment classification,this paper proposes a deep learning model based on collaborative training.For the case where there is a large amount of tagged data in the source domain and a small amount of tagged data in the target domain,the above two models are used to simultaneously extract domain-specific features and domain-shared features.And through collaborative training,the unmarked data of the target domain is effectively utilized.Thereby the performance of cross-domain sentiment classification is improved.