Research Of Aspect-level Sentiment Analysis And Their Application On E-commerce Platforms

With the development of e-commerce platforms,users are accustomed to know the details and quality of products by browsing product reviews.Product reviews include consumer evaluations of various aspects of a product.Therefore,being able to effectively analyze the emotional polarity of various aspects of the commodities in the reviews can not only help users to fully understand commodity information,but also provide the basis for manufacturers to improve product quality.The traditional sentiment analysis task can judge the sentiment polarity of the whole text by analyzing the content of the text,but cannot give the sentiment tendency according to a certain part of the text.As a fine-grained task of sentiment analysis,aspect-level sentiment analysis analyzes sentence text to give the sentiment polarity corresponding to different aspects.Therefore,aspect-level sentiment analysis has direct application significance for analyzing product reviews on e-commerce platforms,and the problems arising from this task are the focus of this paper,and are dedicated to finding solutions and proposing models with better classification effects.The work done in this paper is as follows:(1)In view of the problem that the existing aspect-level sentiment analysis methods combined with word part-of-speech information only extract shallow part-of-speech features and ignore the connection between words and parts of speech,this paper proposes a hybrid attention network based on ELMo(EHAN).Different from existing networks,this model not only combines ELMo with Transformer network to capture the emotional features of textual information,but also fuses the semantic information and part-of-speech information of word parts of speech in the form of attention mechanism to enhance the feature extraction of parts of speech.,enabling the model to learn richer sentiment features for final sentiment polarity classification.The experimental results show that compared with the existing models based on semantic information,the classification effect of the EHAN model is better,which proves that the proposed method can effectively improve the performance of existing aspect-level sentiment analysis methods.(2)In view of the fact that the existing graph convolution network based on the grammar information of the sentence cannot effectively use the edge information of the dependency tree and ignore the specific dependency type,this paper proposes a graph convolution network based method.Multi-attention based on Graph Convolutional Network(MAMGCN).The model adds the dependency type to the grammar graph to help the graph convolutional network obtain more accurate grammar information;secondly,in order to fully obtain the key emotional information in the sentence,the model introduces the grammar attention mechanism and the multi-head attention mechanism.While capturing semantic emotional feature information,it can also learn grammatical effective feature information,thereby improving the classification accuracy of the model.The experimental results demonstrate the effectiveness of the model.(3)In order to apply aspect-level sentiment analysis,this paper further designs a commodity review sentiment analysis system based on the e-commerce platform on the basis of the proposed model,so that the trained classification model proposed in this paper can be deployed to the system middle.Specifically,the Python-based crawler framework Scrapy is used to collect product reviews in multiple fields of e-commerce platforms,and preprocess the data to make the data more standardized and facilitate the model to perform sentiment classification.Then,the system implements the functions of sentiment analysis,result visualization,and polarity comparison based on the network model for users to use.