Citations Sentiment Analysis Based On Classification Of Unsupervised Domain Adaptation

The citation sentiment analysis is to analyze the citations from polarity and function,and mines the potential relationship between citing literature and cited literature.Citations are the descriptions of the authors of citing articles on the work done by cited articles.The sentiment analysis based on machine learning is mainly to convert it into a classification problem,that is sentiment classification.This paper treats both the polarity analysis of citations and the functional analysis of citations as automated classification tasks,ie,predicting the polarity or functional category of each citation by training the classifier.The existing methods for automated classification mainly focus on supervised classification methods,which require a large amount of annotated data.Due to few annotated data in the citation area and the high cost of manually annotated corpora,it is necessary to develop a classification method with less demand for annotated corpora.This paper proposes a classification method based on unsupervised domain adaptation to classify the polarity and function of each citation.The idea of unsupervised domain adaptation is to learn the mapping between the source domain and the target domain when the target domain data is completely unlabeled,and is suitable for situations where the demand for label corpora is small.The classification method proposed in this paper introduces the concept of domain confrontation based on the idea of unsupervised domain adaptation to learn the mapping between source and target domains.The analysis of the polarity of citations is an analysis of the emotional bias of citations.It is divided into three categories:positive,negative,and objective.This paper takes the labeled citation data in the field of computer linguistics as the source domain,and the target domain to be classified is the unlabeled citation data in the biomedical field.The automated classification scheme raised the accuracy of the citation polarity classification result to 90%.The citation function analysis is to identify the motivation and purpose of citation literature cited citations.This paper proposes a ten-category citation function scheme,which is mainly to join the perfunctory category,and manually label the citation feature labels according to the scheme.The automated classification of citation functions also uses an unsupervised domain-adapted classification method.Since most existing studies only focus on polarity analysis or functional analysis,they ignore the internal relationship between citation function and polarity.This article proposes the idea of the mapping relationship between citation function and polarity.The experiment proved that the idea is reasonable,propose the relationship between citation function categories mapped to citation polarity.The unsupervised domain adaptation classification method proposed in this paper has been proved to be effective and feasible after two groups of experiments on the polarity and function of the citation.By analyzing the polarity and function of citations,it is possible to dig deeper the potential correlation between citations and citations.This is the basic work of excavating academic texts.