Research On The Aspect-Level Sentiment Classification Method Based On Structural Relationship Modeling

With the rapid development of the Internet,online consumption has become a trend and has been accepted by majority.Therefore,online comments of various shopping platforms has also begun to show explosive growth.By exploiting the potential emotional tendencies of these comments,enterprises can find their orientation quickly,thereby improve their products and services based on user experience,also gain the trust and favor of consumers and promote the vigorous development of the service industry.Thus,it is of significant commercial and academic value to explore online comment information and to study the emotional motivations behind it.This thesis is mainly devoted to analyzing the various structural relationships that may exist in the text,and discussing the hidden information brought by various structural relationships,constructing two aspect-level sentiment classification models based on structural relationships.Experimenting extensively to prove the effectiveness of the method.The main contents of its research work are as follows:Existing methods have proposed effective models and achieved satisfactory results,but they mainly focus on exploiting local structure information of a given sentence,such as locality,sequentiality or syntactical dependency constraints within the sentence.Therefore,we simultaneously introduce both global structure information and local structure information into the task of aspect-based sentiment classification.In particular,we exploit the syntactic dependency structure as well as sentence sequential information to mine the local structure information of a sentence.On the other hand,we construct a word-document graph using the entire corpus to reveal the global dependency information between words.In addition,an attention mechanism is leveraged to effectively fuse both global and local dependency structure signals.Extensive experiments are conducted on five benchmark datasets,and the results illustrate that our proposed framework GL-GCN outperforms stateof-the-art methods for aspect-based sentiment classification.Previous methods based on GCNs mainly rely on exploring the structure of the syntactic dependency relationships,which would suffer from the sparsity issue.Moreover,these methods usually model one kind of structure information(i.e.syntactic dependency structure)while largely overlooking other kinds of rich structure information between words,such as the consecutive structure of words within a time window,or the cooccurrence structure between words in the entire corpus.To tackle these problems,we extend the GCN model and propose a novel Structure-Enhanced Dual-Channel Graph Convolutional Network(SEDC-GCN).Specifically,we first exploit the rich structure information by constructing a text sequence graph and an enhanced dependency graph,then design a dual-channel graph encoder to model the structure information from the two graphs.After that,we propose two kinds of aspect specific attention,i.e.,aspect-specific semantic attention and aspect-specific structure attention,to learn sentence representation from two different perspectives,i.e.,the semantic perspective based on the text encoder,and the structure perspective based on the dual-channel graph encoder.Finally,we fuse the sentence representations from the above two perspectives and obtain the final sentence representation.The results demonstrate that the proposed approach SEDC-GCN consistently outperforms all competitive baselines.