Research On Enterprise Brand Public Opinion Based On Fusion Model CNN-LSTM

The Internet tertiary industry developing rapidly has accelerated the improvement of mass consumer lifestyles and quality.The arrival of the 5G era is the arrival of business opportunities.Both online and physical enterprises are riding the"east-wind" of big data to accelerate the pace of big data and the Internet of Things.The" China Automotive Industry Development Report" with the theme of"Automotive Industry Reform and Collaborative Development for Future Travel"released in October 2019 pointed out that the traditional auto industry will face ecological changes,and the promotion of auto industry reform must be a technological revolution.At present,the pressure on the domestic economic and social transformation and development environment is increasing,and the structural sudden public opinion factors of the social cycle are increasing.Only by analyzing and researching big data and obtaining data intelligence can we seize the opportunity.Therefore,it is necessary to look at the occurrence of public opinion from a dialectical point of view,conduct in-depth analysis of events that induce public opinion,and adhere to the attitude of "if there is change,then without encouragement",and further improve the management and reform of the enterprise and the brand itself.Although the operation of dictionary-based public opinion analysis is simple,it has high quality requirements for sentiment dictionaries and relatively high labor costs;although machine learning-based public opinion analysis solves the problem of emerging vocabulary,there are problems of isolating feature items and ignoring the connection between contexts;The public opinion analysis based on deep learning has improved the problems in the current text public opinion analysis.The popular models CNN,RNN,LSTM and innovative models have made some contributions to the public opinion analysis,but the public opinion based on the fusion model in the short text side are few analytical studies.This article aims to enrich public opinion analysis in the field of natural language research.Accurate and delicate public opinion models can realize the timely grasp of public opinion dynamics,and enrich neural network algorithms to contribute to the field of deep learning research.This article applies the idea of public opinion analysis based on deep learning to the commercial field of car sales.This article takes the market public opinion triggered by the "Youth Tour" as the starting point,uses the review data of different models on the market obtained by various media platforms and consumer review platforms as the data source,and then influences the brand of the MAZDA brand based on the deep learning model.Public opinion research.The data source field includes the data source platform,the brand to which the data belongs,and the acquisition time:the acquisition time is from July 1,2019 to July 31,2019;the source platform is a consumer platform represented by a blog and a car.The coal body platform represented by the Internet;its brands are the MAZDA brand and its competitors in the market,Lectra 02,BMW X2,Qijun,Rongfang,Honda CRV,and Qashqai.The MAZDA brand review data includes three models of CX-4,CX-5 and Atez.The data source as a whole is the expected set of word segmentation training and Word2vec vectorization;the data consists of two parts.The MAZDA competitive brand data has artificially marked emotional tendencies,and is used as the training set and verification set of the deep learning model at a division ratio of 0.2.The data does not have artificially marked emotional tendencies,and will be used as a test set for deep learning models.On the one hand,it is used to verify the usability of the model,and on the other hand,the branded MAZDA brand data is used to analyze brand influence.After the data is deduplicated and cleaned,Jieba text word segmentation,TF-IDF keyword establishment training dictionary,word2vec establishment word vector,3 establishment token index,then determine the log loss function,Adam optimization function,add Dropou layer to improve generalization ability.After the overall preparation of the model is completed,the CNN-LSTM model created by the TensorFlow+Keras framework and using the CNN-LSTM fusion of the convolutional neural network CNN and the long and short-term memory neural network LSTM is used for emotion classification.The research model combines the advantages of RNN and CNN.It not only considers the sequence information of a sentence but also captures some key words and features in the sentence.The CNN-LSTM model obtains 92.98%accuracy through continuous training.In order to verify the practicability of the model,a comparison experiment was conducted with the relevant single model.The experimental results showed that the single CNN model achieved 89.49%accuracy,and the single model LSTM achieved 92.3%accuracy.None of them reached the accuracy of the fusion model.The effect of fusion model is better than that of pure CNN model and LSTM model.After the model is trained to determine the parameters,sentiment classification is performed on the MAZDA brand data test set.Perform influence analysis on the NAZDA brand data with emotional tendency.Based on the trained CNN-LSTM model,this paper conducts an empirical analysis of MAZDA market review data,analyzes the dynamic communication situation from the time dimension,and divides the public opinion communication process into four periods:birth period,growth period,maturity period,and recession period;then from a cross-sectional perspective Use textrank to extract keywords to analyze the emotional orientation and social concerns of the three MAZDA models.It is found that the three models of CX-4,MX-5 and Atz have been recognized by the nationals with their young sports style and people-friendly handling,but the nationals have strong hard requirements for large spaces and passability.The bottleneck in the development of the Chinese market.MAZDA once placed the Chinese auto market as important as the United States,and its market share has not increased in recent years.This article puts forward three suggestions from the perspective of MAZDA brand positioning and self-development:Advocating multiple communication methods,you can use novel methods with better communication power,such as family dramas;optimizing multiple marketing methods,you can carry out public welfare marketing,pass the brand Values,breaking the prejudice of national cognition;the implementation of a variety of management methods,can cross-border sports music education and other fields of life.