Source-linked AI summary
Deep Sentiment Classification and Topic Discovery on Novel Coronavirus or COVID-19 Online Discussions: NLP Using LSTM Recurrent Neural Network Approach
Hamed Jelodar, Yongli Wang, Rita Orji, Hucheng Huang
TL;DR
The paper asks how public COVID-19 discussions can reveal semantic issues and sentiment relevant to online health communities. It extracts Reddit comments, applies LDA-based topic modeling and an LSTM model, and reports meaningful topics, sentiment analysis, and practical information for understanding COVID-19 concerns. The study is limited to English-language comments collected between January 20 and March 19, 2020.
Problem
The paper addresses the need to analyze COVID-19-related public opinions for semantic issues and sentiment in online health discussions.
Method
The study automatically extracts COVID-19 Reddit discussions, uses LDA topic modeling for semantic extraction, and applies an LSTM recurrent neural network for sentiment classification.
Results
The framework generated meaningful latent topics and sentiment information from COVID-19 comments, with the LSTM model producing better results than several well-known machine-learning methods.
Takeaways & Limitations
Public COVID-19 comments can provide information about people’s concerns and needs and may aid public-health strategies and interventions.
Takeaways & Limitations
The study covers only English-language comments collected from January 20 to March 19, 2020, so results do not represent other languages and may be less timely.
Abstract
from arXiv · showhide
Internet forums and public social media, such as online healthcare forums, provide a convenient channel for users (people/patients) concerned about health issues to discuss and share information with each other. In late December 2019, an outbreak of a novel coronavirus (infection from which results in the disease named COVID-19) was reported, and, due to the rapid spread of the virus in other parts of the world, the World Health Organization declared a state of emergency. In this paper, we used automated extraction of COVID-19 related discussions from social media and a natural language process (NLP) method based on topic modeling to uncover various issues related to COVID-19 from public opinions. Moreover, we also investigate how to use LSTM recurrent neural network for sentiment classification of COVID-19 comments. Our findings shed light on the importance of using public opinions and suitable computational techniques to understand issues surrounding COVID-19 and to guide related decision-making.
1 Introduction
The paper analyzes COVID-19 discussions on Reddit to discover topics and classify sentiment using NLP and LSTM methods. It frames public online opinions as useful evidence for understanding COVID-19 issues and supporting decision-making.
- The study uses automated extraction and NLP topic modeling to uncover COVID-19 issues from public Reddit opinions.
- It presents an NLP framework for extracting meaningful topics from COVID-19-related Reddit comments.
- It proposes an LSTM deep-learning model for sentiment classification of COVID-19 comments, reporting better results than several well-known machine-learning methods.
- The analysis uncovers COVID-19 topics, examines comment sentiment and polarity, and draws on comments from 10 subreddits.
- The findings emphasize public opinions and suitable computational techniques for understanding COVID-19 issues and guiding related decision-making.
2 Related Work
Prior research has applied machine- and deep-learning methods to sentiment and semantic analysis in online health forums. This paper extends that line of work to COVID-19 Reddit comments using topic modeling and an LSTM sentiment-classification model.
- Machine- and deep-learning approaches are established methods for analyzing text content in online health forums.
- Earlier studies examined emotional status, information needs, health-related topics, and clinic-related posts across online forums and health websites.
- The paper identifies its work as the first study to use NLP methods to evaluate COVID-19-related comments from subreddit forums.
- Its proposed approach combines topic-modeling algorithms for meaningful-topic extraction with an LSTM RNN for sentiment classification.
3 Framework Methodology
The framework combines Reddit comment collection and preprocessing with LDA-based topic discovery and LSTM-based sentiment classification. It uses semantic and sentiment analysis to mine COVID-19-related opinions and identify discussed topics.
- Preparing the input data: The framework analyzes COVID-19-related comments from 10 Reddit sub-reddits as input data.Reddit users post and discuss subjects within subreddits, which provide the comment source for the model.
- Removing Noise and Stop-words: Preprocessing removes stop-words and noise from the comments before arranging the dataset for analysis.Stop-word removal also reduces feature-space dimensionality by eliminating common words considered minimally informative.
- Semantic Extraction and COVID-19 Comment Mining: LDA topic modeling with Gibbs sampling extracts semantic content and discovers latent topics among COVID-19-related comments.The model represents comments and words through topic distributions governed by Dirichlet priors.
- Semantic Extraction and COVID-19 Comment Mining: The topic-mining process computes word and document-topic distributions, then recommends and ranks meaningful topics.Algorithm 2 describes probability computation for word distributions and document-level topic distributions, while the recommendation process ranks highlighted topics.
- Deep-Learning and Sentiment Classification: LSTM gates regulate memory retention, new information, and exposure of the internal cell memory.The forget, input, and output gates respectively control previous information, saved information, and exposed internal memory.
- Deep-Learning and Sentiment Classification: An LSTM model classifies sentiment in processed Reddit comments using two LSTM layers and pretrained 50-dimensional GloVe embeddings.The processed comments are converted into fixed-dimension vectors before sentiment assessment.
4 Experiment Details
The experiment analyzed 563,079 Reddit comments with LDA topic modeling and sentiment analysis, then evaluated LSTM and supervised machine-learning approaches for COVID-19 sentiment classification.
- Dataset and topic modeling: 563,079 COVID-19–related Reddit comments were analyzed, with MALLET used to infer latent topics through an LDA topic model.The dataset was collected from January 20, 2020 to March 19, 2020.
- Latent topics: Topics 85 and 18 concerned people and infection, with Topic 85 ranking first and emphasizing terms such as people, virus, and family.The words people and virus had word-weights of 0.1295% and 0.0301%, respectively.
- Latent topics: Topic 63 represented medical staff issues, highlighting hospital, medical, and healthcare with word-weights of 0.0561%, 0.0282%, and 0.0278%, respectively.The topic also included patients, staff, workers, and emergency.
- Latent topics: Topic 30 captured feelings and hopes, where positive words were more prominent than negative words.Topic 93 focused on people, age, and COVID issues, while another topic emphasized quarantine through coronavirus and quarantine.
- Sentiment classification: SentiStrength assigned sentiment scores, which were used to calculate average negative, positive, or neutral sentiment and label five sentiment classes.The five labels were very positive, positive, very negative, negative, and neutral; training and testing sets contained 338,666 and 112,888 comments.
5 Discussion and Practical Findings
The study presents Reddit as a source of public opinion and reports that topic modeling revealed meaningful COVID-19 discussion themes, while the LSTM outperformed several established machine-learning methods.
- Practical value: Reddit comments can provide information about people’s opinions that may be difficult to obtain through traditional manual techniques.The paper describes this as the first study to analyze COVID-19 Reddit comments through both semantic and sentiment aspects.
- Topic findings: 563,079 comments from 10 subreddits yielded meaningful latent topics associated with various COVID-19 issues.The analysis examined dependencies between semantic aspects of user comments and different COVID-19 topics.
- Topic findings: Top-ranked topics contained words related to users’ needs and highlighted discussions on Reddit.A topic is described as a categorical distribution over words.
- Practical value: The framework combined NLP topic modeling with an LSTM sentiment classifier to generate information from COVID-19–related comments.The authors state that these statistical contributions can help determine positive and negative actions in an online community.
- Limitations: The results were limited to English-language comments collected between January 20, 2020 and March 19, 2020, which may affect timeliness and cross-language coverage.The authors note that comments in other languages were not represented.
6 CONCLUSION
The paper applies NLP topic discovery and an LSTM model to analyze the association between COVID-19 sentiment and semantic topics in Reddit healthcare discussions.
- Conclusion: The study identifies itself as the first analysis of the association between COVID-19 comment sentiment and semantic topics on Reddit.Its stated goal was to detect meaningful latent topics and classify sentiment in COVID-19–related healthcare-forum discussions.
- Conclusion: The authors suggest that the results may aid understanding of people’s COVID-19 concerns and needs and support public-health strategies and interventions.The stated application scope includes public health services and interventions related to COVID-19.
Ethical Approval
The study reports compliance with applicable ethical standards and declares no directly related conflict of interest.
- Ethical approval: The procedures followed institutional or national ethical standards and the 1964 Helsinki declaration and later amendments or comparable standards.The statement applies to studies involving human participants.
- Conflict of interest: All authors declared no conflict of interest directly related to the submitted work.