Research And Application On Aspect-based Text Sentiment Analysis Algorithm Based On Transfer Learning

In daily life,people post a large amount of review information to various social networking sites and shopping sites through various channels,expressing their positive or negative,or supported or opposing emotions.Sentiment Analysis(SA)technology is a process that helps users quickly obtain,organize,and analyze these review information through computer tools.Aspect-Based Sentiment Analysis(ABSA)is a type of sentiment analysis that extracts opinions expressed on different aspects of an entity and is a refined sentiment analysis model.Aspect extraction is one of the important steps in aspect-based sentiment analysis model.This paper carries out research on aspect extraction based on transfer learning of product review data.The main research contents include the following:First,this paper proposes a Transfer Learning Algorithm for Aspect Extraction(TLAE).The algorithm combines the idea of Label Propagation Algorithm(LPA)with a transfer learning framework.The first step is to perform a parsing analysis on the original review data to extract the linguistic features of the text review data in all fields,in order to reduce the original feature space from high-dimensional to low-dimensional.In the second step,the data representation graph is constructed to represent all domains data.The graph nodes are the aspect labels of the source domain data,the aspect candidate labels of the target domain data and the language features of all domains data.The edges of the graph are the connection lines between language feature nodes and aspect nodes.The weight of the edge is the frequency between the aspect nodes and the language feature nodes.The third step is to implement aspect label propagation in the source and target domains based on the data representation graph to classify the aspects of the target domain data.The algorithm is tested on English and Chinese datasets in different domains,and is compared with other algorithms which are used to extract aspects.The results show that the lower the correlation between the source domain and the target domain,the higher the classification accuracy of the TLAE algorithm,which is 1.31%-3.96% higher than that of the optimal algorithm.This algorithm has an absolute advantage in classification time.Second,this paper constructs an opinion mining system based on TLAE algorithm for aspect-level opinion mining of product review data.The system extracts texts that contain evaluation entities and entity aspects,then mines comment opinions on different aspects of the entities and visually analyzes them.It can give consumers and businesses a more comprehensive understanding of all aspects of the product to help them make better decisions.