Analysis Of Public Opinion Information Based On Big Data Of Wine Industry

In recent years,major liquor companies in China are actively working on digital transformation and upgrading.In this process,in order to deeply tap user preferences and grasp industry hotspots,it is necessary to fully analyze relevant public opinion informa-tion on the network.However,there is no clear category label for comments,news,and developments on the Internet,which increases the difficulty of public opinion analysis.How to effectively analyze a large number of online public opinion texts is a problem faced by various wine companies.The traditional public opinion information analysis is mostly based on supervised learning models and the models are independent of each other and lack of contact.This paper proposes new solutions to the problems of a large sample size of network public opinion data and difficulty in labeling.This article mainly carried out the following research:1)This paper uses a semi-supervised text classification model based on graph con-volution to classify public opinion information.The public opinion information collected by web crawlers is messy,and first of all,it is necessary to classify topics according to certain standards.Due to a large amount of online public opinion data and the high cost of annotated data,the annotated data accounts for a relatively small amount,so the traditional supervised text classification algorithm is not ideal in this scenario.In order to solve this problem,this paper proposes a semi-supervised text classification algorithm SS-GCN based on graph convolutional network,which transforms text classification tasks into graph classification tasks,implements semi-supervised learning,and the classifica-tion accuracy rate is 0.896.The comparison experiment with the classic text classification algorithm confirms the feasibility and efficiency of semi-supervised learning based on graph convolution.2)This paper improves the Single-Pass algorithm by adding initial clustering and an inverted index.After the wine industry,public opinion data is classified by subject,fur-ther fine-grained topic detection is required to obtain specific events under different topics.The improved Single-Pass algorithm takes the embedded representation of the document generated by the SS-GCN model as input and realizes dynamic topic detection while re-ducing the amount of calculation and improving efficiency.The improved algorithm CHI index and DBI index are 14677 and 0.508,respectively,which are superior to the tradi-tional clustering algorithm,and the clustering speed is significantly accelerated,and the running time decreases by 30.63)Another important task of public opinion analysis is sentiment classification.This paper proposes the Attention-BiGRU-CNN model.Sentiment classification can reflect the network users' emotional tendency towards an event,while document-level emotion classification is not a simple sentence sentiment superposition.Attention-BiGRU-CNN is a sentiment classification model based on two-way GRU and CNN based on attention.It integrates the current mainstream technology attention,two-way GRU,and CNN in the field of NLP.At the same time,it draws on the CBAM module in the field of computer vision to perform attention.In order to solve the problem of sentiment labeling,senti-ment dictionaries are used to calculate the sentiment scores of documents,and the results with higher confidence are selected as tag data.Attention-BiGRU-CNN model emotion classification accuracy rate is 0.766,which is a certain improvement compared to other models.