Research And Implementation Of Fine-grained Sentiment Analysis And Visualization For User Reviews

"With the explosive growth of digital text data,such as social media and online comments,sentiment analysis and visualization systems have become crucial tools in various fields,including digital social science,business,and opinion research.Sentiment analysis is a natural language processing technique designed to automatically identify sentiment in text.This includes sentiment intensity,sentiment polarity,and sentiment themes,which help to analyze sentiment information in text data.Visualization systems,on the other hand,present data visually,making it easier to observe trends and relationships in the data.However,research on sentiment analysis algorithms is influenced by the quantity,quality,and scope of the datasets.Currently,publicly available datasets primarily focus on restaurants and entertainment,leaving gaps in other areas,including the tobacco industry.This is particularly problematic given the rapid growth of tobacco consumption.From 2000 to 2022,the average annual growth rate of domestic tobacco consumption reached 10.52%.Despite the significance of the tobacco field,there is limited research on it.To fill this gap,this paper aims to analyze the sentiment tendency of online cigarette reviews through a text sentiment analysis visualization system.By visually displaying consumers’evaluation and satisfaction,this study seeks to provide more realistic and effective research assistance in the tobacco field.Ultimately,this research will contribute to a better understanding of consumer demand and product usage experience in the tobacco industry.This paper focuses on four key areas of research.This paper introduces a novel multi-category dataset in the tobacco domain that incorporates a priori knowledge of sentiment features that are well-suited for deep learning.This dataset includes tobacco reviews that cover various aspects such as taste,flavor,price,packaging,quality,and more.Furthermore,each aspect is classified into three sentiment polarities-positive,neutral,and negative-making it suitable for both coarse-grained and fine-grained sentiment analysis tasks.By providing a comprehensive and well-structured dataset,this work fills the gap in the existing public datasets in the field of sentiment analysis and serves as a catalyst for further research in this domain.This paper presents a solution to the challenge of multiple sentiment polarities in review data texts by proposing the theory that sentence polarity is not only determined by its content but also by the sentiment intensity of its associated aspects.To address this,we have designed and implemented a knowledge-driven multi-entity coarse-grained sentiment analysis model,EHCRNN,specifically for the tobacco domain.Our model effectively extracts a priori knowledge features from pre-trained models and deep learning networks to synthesize and analyze multi-polarity data in tobacco domain,providing accurate sentiment polarity results for online tobacco reviews.To validate our approach,we conducted experimental tests on both new tobacco datasets and publicly available datasets.Results show that our model outperforms all existing mainstream methods.Additionally,we conducted ablation experiments and hyperparameter sensitivity experiments,further demonstrating the effectiveness of our model.Our work also contributes a new multi-category tobacco domain dataset containing a priori knowledge of sentiment features suitable for deep learning,which fills the gap of existing public datasets in this field and promotes sentiment analysis research in the tobacco domain.Aiming at the problem that fine-grained aspect and sentiment correspondence cannot be achieved without fine-grained annotated data,this paper proposes the construction of SenEM,a self-supervised finegrained sentiment analysis model that fuses sentiment lexicon and deep learning.Firstly,a sentiment lexicon exclusive to the tobacco domain is constructed using the LDA algorithm,which includes fine-grained aspects of tobacco reviews in addition to the basic sentiment lexicon to build a lexicon of feeling words and sentiment words.Then,a fine-grained sentiment analysis model is developed based on the sentiment lexicon and deep learning model,which solves the problem of carrying out fine-grained sentiment analysis without fine-grained annotation in the dataset by selfsupervision,providing aspects involved in the text and their sentiment polarity.Based on the algorithm model proposed above,a tobacco review sentiment analysis visualization system was designed and realized,and the relevant academic research content was implemented in a systematic form,and current smokers reviewed tobacco reviews of different brands and specifications.Emotion visualization display,using visual tools such as charts,word clouds,and scatter plots to present data to help users better understand and analyze emotional information.These visualization tools can present data visually to help users better understand data trends and relationships.At the same time,control the operation convenience and response speed of the system.Provide real and effective help for the tobacco industry to understand consumer needs and improve product experience.