Fine-grained Opinion Mining Based On Global Dependency And Its Application In Master's Academic Dissertation Evaluation

In the age of information explosion,all kinds of information are presented to the public through different platforms.Review is an effective way to reflect real information and feelings,both in the consumer and professional fields.Reviews can be used to understand users' needs and improve them.In professional fields such as paper reviews and expert review,reviews can help students better complete their studies and improve their academic level.The various characteristic information in reviews is considered to be a fine-grained opinion,which can be mined to help improve the product and make decisions for users.Fine-grained opinion mining is a hot topic in the field of data mining and natural language processing,and more and more researchers attach importance to it.This paper makes an in-depth study of fine-grained opinions in reviews.A fine-grained opinion mining model based on global dependency is proposed from the perspective of Chinese semantics,which mainly solves the task of evaluation target recognition and sentiment classification.Different from the general research on fine-grained opinion mining,the fine-grained opinion mining model proposed in this paper is a linkage model,which can complete the fine-grained opinion extraction,attribute classification and sentiment classification of reviews.The main work of this paper is reflected in the following aspects:(1)In this paper,the structure and grammatical information of sentences is studied and the concept of global dependency is put forward.By taking advantage of the sequential modeling of long short-term memory network,the relative long short-term memory network is proposed in combination with the relative syntactic distance and syntactic dependency.Besides,based on the relative long short-term memory network,a global dependency between words is constructed from time and space.(2)In this paper,the tasks of evaluation target extraction and evaluation object classification in fine-grained opinion mining are studied,and a multi-task learning model based on feature interaction is proposed.Global features are obtained by modeling the whole sentence.Local features are obtained by modeling global dependencies among words.Then,the interaction between global features and local features is carried out by using the correlation and difference of evaluation target extraction and attribute classification tasks.Finally,by sharing interactive features in a highly generalizable multi-task model,two tasks are completed in different ways.(3)In this paper,the sentiment classification task is studied,and a sentiment classification model based on target relational graph and double attention network is proposed.In order to analyze the sentiment of the evaluation target,a model with double attention network characteristics and target relational graph characteristics was formed by constructing the target relational graph with the target as the core and the global dependency relationship as the edge,so as to obtain the sentiment information of the evaluation target and the overall review.The target relational graph can focus on the relationship between the target and the context.The double attention network can model the relationship between the target and the context,and the relationship between the target and the overall review respectively.In this paper,the multi-task learning model based on feature interaction is combined with the sentiment classification model based on target relational graph and double attention network as a complete fine-grained opinion mining model,that is,the fine-grained opinion mining model based on global dependency.In order to prove the practicability of this model,in view of improving the expert reevaluation in the current master's thesis evaluation proposed by Shanghai Academic Degrees Office,this model is applied and researched in the master's thesis evaluation.Through fine-grained opinions mining on expert review text to support and improve the expert appraisal system.respectively is to analysis comprehensive opinion of the unqualified papers to find out the specific problems of the unqualified papers?to provide decision support for expert appraisal?to evaluate the consistency of expert review opinions,so as to improve the quality of the expert review and supervise the experts appraisal behavior.