Research On Fine-grained Sentiment Analysis Algorithm Based On Aspect Term And Opinion Term

Sentiment analysis is an important branch of natural language processing.From the granularity of the processed text,sentiment analysis can be divided into coarse-grained sentiment analysis and fine-grained sentiment analysis,and coarse-grained sentiment analysis can be further divided into chapter-level sentiment analysis and sentence-level sentiment analysis.Sentiment analysis and fine-grained sentiment analysis are sentiment analysis that takes aspect term as the processing object.Fine-grained sentiment analysis can be divided into three sub-tasks in terms of task requirements:aspect term extraction,opinion term extraction,and sentiment polarity classification.In the previous fine-grained sentiment analysis methods,most of the research methods directly embed the information of the aspect term into the sentence representation,so as to learn the relevant sentiment features of a specific aspect term,which leads to the lack of specific aspect term.The capture of the emotional relationship between words in terms of aspects largely ignores the impact of emotional relationships between different aspects of words in the sentence on the results of sentiment analysis,and most research methods do not consider irrelevant words or interference in the text of the comment sentence.The impact of words extracted from correct aspect term on feature extraction.In addition,most of the research methods do not consider the semantic relationship between these two sub-tasks when they perform the task of opinion term extraction and aspect term extraction,and they have a guiding role for each other.Aiming at the above problems,this paper designs corresponding feasible solutions.The main work of this paper is as follows:(1)This paper designs a graph convolutional neural network model GCN＿C that combines aspect term and the relationship between aspect term.First,construct a common dependency graph for each sentence on the dependency tree,refine the graph by considering the syntactic dependencies between the context words and aspect term,and obtain the graph G of the specific aspect term,using the graph G and the corresponding words The embedding matrix constructs a graph convolutional neural network centered on aspect term,and then,in order to extract the sentiment relationship between aspect term and other aspect term,a graph convolutional neural network that considers the relationship between aspect term is constructed.The experimental results on the official benchmark datasets of Restaurant 2014 and Laptop 2014 show that the GCN＿C model designed in this paper has further improved the effect of sentiment analysis compared with other models.(2)This paper designs a multi-task learning network model M＿NET that combines aspect term information and opinion term information.First,use the textual semantic information of aspect term and opinion term jointly to mark all possible aspect and opinion relations,and then use the mutual indication between different opinion factors to design an inference optimization strategy,further optimize the labeling of aspect opinion relations,and finally mark the final markup.The results of the decoding operation are performed to obtain more accurate jointly extracted aspect opinion pairs and opinion triples.The experimental results on the official benchmark datasets of Restaurant 2014,Restaurant 2015 and Laptop 2014 show that the M＿NET model designed in this paper has further improved the effect of sentiment analysis compared with other models.