| Fine-grained sentiment analysis is an important task in natural language processing,and its purpose is to automatically identify the sentiment information expressed in the text.With the widespread popularity of the Internet and social media,users post comments,messages,and other texts more and more frequently.Generally,these texts contain sentiments expressed by the authors,so a sentiment analysis can help businesses,governments,and scholars gain a deeper understanding of market trends,user needs,and social sentiment.Therefore,fine-grained sentiment analysis has a wide range of applications.To solve the fine-grained sentiment analysis task completely,the researchers proposed the task of Aspect Sentiment Triplet Extraction(ASTE)as a solution,which is the process of extracting the sentiment triplets from the text,including aspect terms,opinion terms,and sentiment polarities.This paper presents the work done on the ASTE task,which includes:(1)This paper proposes an SSJE(Span-sharing Joint Extraction)framework for ASTE tasks,which provides a relatively complete solution for fine-grained sentiment analysis.SSJE uses joint extraction to identify the sentiment polarity of aspect terms,opinion terms,and aspect sentiment polarity at the same time,which effectively avoids the error propagation encountered by traditional pipelining methods in different sub-tasks.In addition,by sharing the span of all possible aspect terms and opinion terms,SSJE also solves the problems caused by aspect terms and opinion terms composed of multiple words,and can effectively deal with the one-to-many and many-to-one relationship between aspect terms and opinion terms.Furthermore,the graph neural network is used to enhance the span of semantic representation by integrating the dependency information from the sentence dependency tree.This can effectively improve extraction performance.Finally,this study conducted comprehensive experiments on two public datasets,demonstrating that the proposed framework outperforms all other approaches in terms of performance on both ASTE and AOPE tasks.(2)Based on the research above,a joint opinion entity extraction system is designed and implemented in this paper.This paper introduces the design of the system from two dimensions: logical architecture and process architecture.In the process of system design,to further improve the expandability of the model,this paper constructs a new corpus through data acquisition and preprocessing,and carries out feature extraction and model training for this corpus,so that the model can identify the sentiment triplets that exist in Chinese comments.In the implementation part of the system,to facilitate users use,this paper uses B/S architecture to complete the design and implementation of the system and builds a visual interface.Finally,the function test and performance test of the system are carried out to improve the stability of the system. |
