Research On Aspect Extraction Method For User Online Comments

Aspect extraction is one of the main tasks of fine-grained sentiment analysis,which aims to identify the person or thing being rated in online comments.At present,aspect extraction methods based on deep learning are mainly divided into two categories.One is to first extract the aspects in the comment texts in a pipelined way,and then judge the sentiment polarity of the comment texts based on the given aspects;the other category is to extract the sentiment polarity corresponding to aspects,opinions and aspects in the comment texts in a unified way.For this reason,this paper takes user online comment data as the research object,combines natural language processing and deep learning related technologies,and conducts research on aspect extraction and sentiment analysis from the perspectives of single-task and multi-task.The specific research work is as follows:(1)Aiming at the problem of single-task aspect extraction,an aspect extraction model based on double-embeddings and sequence generation architecture is proposed to extract aspects from comment texts.The model is mainly composed of a presentation layer,an encoding layer and a decoding layer.The presentation layer uses double-embeddings technology to enhance the representation of text to make up for the lack of semantic expression capabilities of traditional word embeddings;the encoding layer uses Bi GRU to capture the overall meaning of the sentence,and the decoding layer is used to generate label sequences.This paper conducts experiments on two public datasets,and the experimental results verify the effectiveness of the model.(2)Aiming at the problem of single-task sentiment analysis,an interactive graph attention network model for aspect-level sentiment analysis is proposed,which aims to judge the sentiment polarity of comment texts on the basis of given aspects.The model models the positional relationship between aspects and contexts by introducing positional embeddings to measure their semantic relevance;and by introducing a graph attention network to model the syntactic relationship between aspects and contexts,to provide corresponding syntactic constraints for aspects and contexts,the problem of long-distance dependencies is effectively alleviated;and by introducing an interactive attention network to model the semantic relationship between aspects and contexts,it is used to assign reasonable attention weights to aspects and contexts.The experimental results on three public datasets show that compared with other existing models,the accuracy and macro-average F1 value of our model are significantly improved.(3)Aiming at the multi-task aspect extraction problem and sentiment analysis problem,an aspect-opinion-sentiment triple extraction model based on graph convolutional neural network is proposed to extract aspect-opinion-sentiment triples in comment texts.The model firstly enhances the representation of the comment texts by means of double-embeddings technology to make up for the lack of semantic expression ability of traditional word embeddings;then uses Bi LSTM to learn the semantic feature representation of sentences;Then uses graph convolutional neural network to model the syntactic relationship between aspects and opinions to provide direct or indirect syntactic constraints between aspects and opinions;then the pairing relationship between words is detected through the inference strategy and the prediction results are output,and finally the triplet is output through the decoding algorithm.In order to evaluate the performance of the model,this paper conducts experiments on two public datasets,and the experimental results verify the effectiveness of the model in this paper.