Fine-grained Sentiment Analysis Of Clothing Reviews

With the rapid development of Internet technology and the continuous improvement of ecommerce platform,the number of groups who choose to buy clothing online is increasing year by year.At the same time,a large number of clothing review texts were produced.The review text expresses the consumer’s subjective emotional tendencies towards purchasing goods.Effective analysis of these clothing reviews will not only help businesses improve product quality and service level,but also help consumers make purchase decisions.How to quickly and effectively mine consumers’ opinions on products and their attributes from the massive review texts has become a hot issue in the field of sentiment analysis.Traditional text sentiment analysis is mainly to judge the sentiment tendency of texts and sentences.However,with the increasing market competition,such traditional coarse-grained sentiment analysis methods are difficult to effectively obtain the increasingly prominent individual needs of consumers.Fine-grained sentiment analysis aims to obtain the emotional tendency information of various attributes of commodities.It has important application value for helping merchants and consumers to understand commodity information more fully,and has attracted more and more attention from researchers.However,most of the existing research results of fine-grained sentiment analysis are related to electronic products,automobiles,restaurants,etc.Since there is no publicly organized clothing review data,there is less fine-grained sentiment analysis in the clothing field.In addition,there are still many difficulties and challenges in fine-grained sentiment analysis,such as the lack of corresponding evaluation objects(entities or attributes)for evaluation words in the review text,and the polysemy of Chinese words.This paper focuses on the corpus construction,attribute extraction,attribute sentiment classification and implicit attribute recognition tasks in the fine-grained sentiment analysis of clothing reviews.The main work of this paper is described as follows.Firstly,aiming at the basic problem of lack of clothing reviews data set in the research of finegrained sentiment analysis,we collected 12983 clothing reviews of a clothing brand on T-mall website.After a series of preprocessing operations,9640 valid comments were obtained.According to different tasks in fine-grained sentiment analysis,different annotation schemes are designed for data labeling to build a clothing review corpus for fine-grained sentiment analysis.The proposed tagging scheme takes the character as the unit to label,which effectively avoids the inaccurate word segmentation problem in natural language processing.Secondly,for the task of attribute extraction and attribute sentiment classification in finegrained sentiment analysis,a unified model TE-Bi GRU-CRF which applies a unified tagging scheme is proposed.The model is composed of transformer encoder,Bi GRU network and CRF network.It makes full use of the strong inter word relation capture ability of transformer encoder and the context feature extraction ability of Bi GRU network,and then learns label constraints through CRF network.The transformer encoder makes up for the Bi GRU network’s inability to perform parallel computing and difficult to process long sentence sequences,thereby effectively improving the prediction effect of the model.The experimental results show that compared with the traditional Bi GRU-CRF model and TE-CRF model,the F1 value of TE-Bi GRU-CRF model designed in this paper is increased by 1.12% and 6.3% respectively,which proves the feasibility and effectiveness of the model.In addition,this method adopts a unified annotation scheme to solve the task of attribute extraction and sentiment classification in the fine-grained sentiment analysis of comments in an end-to-end method,which has higher practical application value.Finally,this paper proposes a model Bi GRU-IC-CRF for the implicit attribute recognition of online clothing product reviews,which is used to identify product attribute names and evaluation word sets and their internal corresponding relationships from online clothing reviews without clear product attribute names.The basic architecture of our model is used to integrate the IC module on the basis of the Bi GRU-CRF model,which further improves the recognition effect of the model.The IC module is mainly based on a gating mechanism to maintain the consistency of corresponding attributes within the opinion words.We employ the Bi GRU to obtain the contextual information of the data,which effectively solves the polysemy problems in Chinese and the problem of emotion words modifying different attributes in different contexts.The results show that this method has a higher recognition rate,F1 value reached 85.48%.Compared with Bi GRU,Bi GRU-IC and Bi GRUCRF method,the F1 value increased by 4.15%,3.98%,1.16% respectively.This method is not only applicable to the implicit attribute recognition in clothing review,but also helpful to other fields implicit attribute recognition of product reviews.