Research On Aspect Level Sentiment Classification Based On Deep Learning

In recent years,with the development of Internet technology,people share their lives and express their opinions on various things on the Internet.And user reviews on various social networking platforms and e-commerce platforms have exploded.How to extract users' specific sentiments about a product or service in a certain aspect fro m many reviews is of great practical significance for merchants to improve their own products and services,as well as to guide other users to purchase goods.Aspect-based sentiment classification aims at identifying the sentiment polarity of the given aspect in a sentence,which analyze the sentiment information of different attributes or aspects in text reviews from a more fine-grained perspective to gain a more in-depth and detailed understanding of goods and services.This thesis focuses on aspect-based sentiment analysis and the main research contents and contributions are as follows:(1)Most of the previous methods are based on Long Short-Term Memory Network(LSTM)and attention mechanisms,which largely rely on the semantic correlation between aspects and contextual words in the modeled sentence,but ignore the syntactic information in the sentence.To tackle this problem,a method based on graph convolutional networks and attention mechanism was proposed,which not only models the semantic information between words in a sentence using a bidirectional long short-term memory network(Bi LSTM),but also uses graph convolutional networks(GCN)to transfer information about the syntactic relationships.Finally,the attention mechanism is used to modeled the relationship between context and aspect for sentiment classification.(2)Similar to the previously proposed method,several existing methods also use graph convolutional networks to convey syntactic information,whose main idea is that reduce the distance between aspect and opinion words through a graph convolutional approach based on the dependency tree and transfer information through the syntactic relations between sentences.On the one hand,this approach is highly dependent on syntax parsing tools,which inevitably leads to an accumulation of errors if parsed incorrectly.On the other hand,the existing syntax parsing tools cannot parse long sentences well,and they cannot help with some special sentences,such as double negatives and rhetorical questions.To make full use of the syntactic dependencies of sentences while alleviating the problems due to parsing errors,an interactive attention graph convolutional network(IAGCN)is proposed that deeply models the semantic and syntactic information of sentences and aspect.Firstly,IAGCN starts with a bi-directional long short-term memory network(Bi LSTM)to capture contextual semantic information regarding word orders.Then,the position information is introduced and put it into the graph convolutional network to learn the syntactic information.After that,aspect representation is obtained through mask mechanism.Finally,the interactive attention mechanism is used to interactively calculate and generate the aspect-specific contextual representation as the final classification feature.Through this complementary design,the model can obtain a good contextual representation that aggregates the aspect target information,and is helpful for sentiment classification.Experimental results show that the model achieve a good performance on multiple datasets.(3)Finally,the workflow of the whole experiment was visualized,example sentences with each characteristic were selected from the dataset,the difficult ies of aspect-based sentiment classification under different sentence forms such as ellipsis,negation and interjection were systematically analyzed,and an aspect-level sentiment classification system platform was designed and built to verify the usability of the model designed in this thesis.