Sentiment Analysis And Reaearch Based On Aspect Level

With the rapid development of e-commerce,more and more consumers take the review information of e-commerce platforms as an important reference standard for consumption choices.According to statistics,the amount of review is increasing at a rate of tens of millions every day.Analyzing these reviews is great benefit to consumers and platforms,and it also promotes the development of aspect level sentiment analysis.Aspect level sentiment analysis is an important task,the purpose is to complete the judgment of the sentiment words polarity of multiple aspects in the text by learning the context information.After in-depth analysis of the existing aspect level sentiment classification algorithm based on the attention mechanism,this paper found the following problems.First of all,the existing attention mechanism lacks consideration of the whole context in the weight distribution of aspect terms and sentiment words.In particular,the deep learning model is directly used of for the analysis of the entire text leads to an imbalance in the weight distribution of aspect terms and sentiment words.Secondly,the existing algorithms are not lacked the use of prior knowledge,but also the phenomenon of polysemous sentiment words has not been resolved.In response to the above problems,the main research contents of this article are as follows:(1)An aspect level sentiment analysis algorithm combines multi-self attention mechanism and sentiment dictionary.The algorithm combines the deep learning model with the sentiment dictionary to extract aspect terms and sentiment words respectively.In the aspect terms extraction,the multi-self attention mechanism is used to jointly model the aspect term category and the context to capture the weight of a given aspect category to form an aspect term matrix in the context.In the sentiment word extraction and classification,using the sentiment dictionary to extract the sentiment word.In order to solve the broadness problem of the basic sentiment dictionary,it is proposed to use phrase structure grammar to extract sentiment words in the text to expand the traditional sentiment dictionary.Convolutional neural network is used to realize the interaction between aspect terms and sentiment words.The algorithm not only can directly extract sentiment words and improve the accuracy of classification,but also the running speed of the model has been improved.(2)An aspect level sentiment analysis algorithm is based on integrating aspect-aware interactive attention and emotional position-aware.The algorithm mainly includes two parts: interactive attention and emotional position-aware.Interactive attention is mainly composed of global attention and local attention.Global attention is to extract the whole features,and local attention is to adjust the weight of aspect words in the context based on the given aspect word category.The position-aware fusion mechanism is to use location information to form a two-tuple of<aspect words,sentiment words>,which solves the problem of polysemous sentiment words.The algorithm not only combines context to classify the sentiment words,but also greatly improves the accuracy of sentiment words classification.