Identification Of Sentiment Features In Academic Reference

Citations are ubiquitous in scientific literature and are an important part of the paper.The author cited the research results of others while writing the paper.On the one hand,the author showed his attention and interest in the achievement.On the other hand,he also incorporated or compared other people's thoughts and methods in his own research results.Nowadays,citation analysis has become an important research field in library and information science,playing an important role in the prediction and analysis of scientific development trends,scientific achievements,and project evaluation.The sentiment analysis of citation contexts as an important part of citation analysis reflects the subjective emotions of citing authors and the influence of cited papers on citing papers.The traditional citation analysis only fo cuses on whether there is a citation relationship between the two documents and the number of citations in a document,and highly cited papers are considered to have strong academic influence.However,high cited does not mean that the paper has a positive influence on citing papers.Some citations may be negative,critical or neutral.Therefore,citations for differe nt emotions should be treated differently.Then,the sentiment analysis against the context of citations is particularly critical.Sentiment analysis needs of different text types are growing,but sentiment in the scientific literature are less studied and citation emotions are one of them.Citation sentiment ana lysis is a research topic with applied value.It can help researchers identify existing problems or undiscovered problems,as well as gaps and shortcomings in current research methods.In the literature retrieval task,the quality of the papers detected can be improved by identif ying negative references.In citation contextual sentiment analysis,the viewpoint holder is the author of the citing document.The evaluation object can be the method,research viewpoint,conclusion or experimental data of the cited paper,which is the reference object of the citing paper.Early scholars believed that the adjectives determined the emotion of the text and there was a certain correlation between the two adjectives connected by the conjunction.At present,there are relatively mature open emotional dictionary resources.Through heuristic methods,these emotional words can be used as seed vocabulary to expand in large-scale corpus to obtain a large-scale dictionary.Compared to product reviews and other UGC texts,scientific and technical terms,such as:state of the art,are also meaningful units in the scientific paper's sentiment analysis.Therefore,constructing a domain emotional dictionary is the focus of research.The research focus of this paper is the detection of reference emotions in scientific lite rature.Through the analysis of the characteristics of scientific and technological literature writing,the author divided the study into three sub-tasks and provided a new large-scale cor pus for each task.First of all,I created a citation emotion corpus with 3,085 negative citation sentences,15441 positive citation sentences,and 70758 citation sentences with no subjec tive sentiment.Secondly,in the sentiment analysis of scientific research texts,the granularity has great significance.Therefore,based on the annotation of emotional categories,this paper also manually annotated the corpus of reference objects.Through the establishment of the reference object recognition model and the mining of citations evaluation object,so as to realize the fine-grained citation emotion analysis.The emotion of scientific texts is generally subtle,so we further conducted comparative sentence recognition experiments to enrich the research of citation emotion analysis.Based on the statistical analysis of the reference position,valuable information in different chapters of the paper can be revealed.And we train a chapter structure function brake recognition model to automatic annotation on different raw materials.