Canada

STEM Fellowship Inter-University Big Data Challenge 2021

Are You Spreading Misinformation? Analyzing How Influential Twitter Users Contribute to the Spread of COVID-19 Information

technical paper

STEM Fellowship hosted the world’s first infodemiology themed hackathon through the 2021 Undergraduate Big Data Challenge, culminating in the national finale event of the Big Data Day. The top 13 teams from across Canada defended their work, and were able to showcase their solutions to a national audience. The winning teams won cash prizes, the opportunity to publish their manuscripts in the STEM Fellowship Journal, and internship opportunities with our partners.

The infodemic created by the COVID-19 pandemic has created several societal issues, including a rise in distrust between the public and health experts and even a refusal of some to accept vaccination; some sources suggest that 1 in 4 Americans will refuse the vaccine [1]. This social concern can be traced to the level of digitization today – particularly in the form of social media. As social media was the most significant contributing factor to the spread of misinformation, the team decided to examine infodemiology across various text-based platforms (Twitter, 4chan, Reddit, Parler, Facebook, and YouTube). This was done by utilizing a sentiment analysis to compare general posts with key terms flagged as misinformation (all of which concern COVID-19) to determine their variety. In gathering the datasets, both APIs and also pre-existing data compiled by standard scientific third parties were used. It was found that in some cases, misinforming posts can have up to 92.5% more negative segment skew compared to accurate posts. From this, the novel Plebeian Algorithm is proposed, which utilizes sentiment analysis and post popularity as metrics to flag a post as misinformation. This algorithm diverges from that of the status quo, as the Plebeian Algorithm uses a democracy process to detect and remove misinformation. A method was constructed in which content deemed misinformation to be removed from the platform is determined by a randomly selected jury of anonymous users. This not only prevents these types of infodemics, but also guarantees a more democratic way of using social media that is beneficial for repairing social trust and encouraging the public’s evidence-informed decision making.



The Plebeian Algorithm A Democratic Approach to Censorship and Moderation

Medical implants are now an integral part of public health in the United States. However, the last few years have brought several controversies regarding their regulation, considering the large number of deleterious effects reported by patients. Some authors even denounce the lack of rigour and transparency in the process of marketing medical implants as well as in the reporting of data. This article aims to address these shortcomings by introducing the American Database for Medical Implant Transparency (ADMIT), a database integrating data from multiple governmental  sources to characterize the entire process by which implantable medical implants are approved by the FDA. After exploring the potential of ADMIT through preliminary analyses, it will be published on an open access platform. Ultimately, it is hoped that this initiative will stimulate research into the real negative effects of gaps in medical implant regulation while providing an informative tool for patients seeking information about them.


Introducing ADMIT : A First Step in Uncovering FDA’s Deceit

During the COVID-19 pandemic, huge amounts of new information from many different sources is made available to us each day. This pandemic has been worsened by this associated ""infodemic"". As individuals are becoming increasingly reliant on the internet as a source of information. This ultimately leads to misunderstandings of public health orders or even indirectly affects public policies. While it is critical to identify misinformation and unreliable news sources online, this task becomes increasingly challenging with the tremendous amount of information released each day. We propose the usage of natural language processing (NLP) and machine learning (ML) techniques to build a model that can be used to identify unreliable articles online. We first processed the ReCOVery dataset to obtain 2039 English news articles tagged with COVID-19 keywords from January to May 2020 which are labeled as reliable or unreliable. Data exploration was done to determine major differences between reliable and unreliable articles. We found that reliable news articles had a higher proportion of neutral sentiment while unreliable articles had a higher proportion of positive sentiment. Additionally, our analysis demonstrated that reliable articles had more engagement on social media on terms of tweets, retweets and likes. We built a multi-input deep learning model called Deep Reli using sentiment and body text of articles to classify the reliability. DeepReli was evaluated to achieve a 93% accuracy, 94% sensitivity, 92% specificity. This paper identifies novel differences between reliable and unreliable articles as well as builds a model using state-of-the-art deep learning techniques. We aim to be able to use our findings to help researchers and the public audience more easily identify misinformation and unreliable media in their everyday lives. Future work by researchers could include creating a larger multi-input model which includes features such as geographical location as well as me-series data of news articles. Additionally, it would be valuable to compare the efficacy of the model on other media such as social media posts or headline as well as with different news articles such as those discussing major political events.

Data Exploration and Classification of News Article Reliability: A Deep Learning Study

Vaccination of the general public is a key step towards reducing the spread of COVID-19 and achieving herd immunity [1]. However, vaccine hesitancy is one of the greatest hindrances to mass vaccination campaigns. As such, being able to understand the factors that affect vaccine sentiment amongst the public is crucial to being able to manage the spread of COVID-19. Previous studies have highlighted the relationship between tweet sentiment and public sentiment, and have also found that stock prices are reflective of public opinion towards a particular company. Since Pfizer, Moderna and AstraZeneca are the primary vaccine producers, tweets relating to the vaccines produced by these companies as well as the stock prices of these companies are likely good indicators of public sentiment towards the COVID-19 vaccines. To identify and quantify the influence of pandemic events that have affected people’s sentiments towards the Pfizer, Moderna, and AstraZeneca vaccines the most, Twitter sentiment analysis as well as a stock price analysis for these companies was performed. A data set of 4,093,081 tweets authored by 1,708,892 unique users which contained key phrases or hashtags related to the AstraZeneca, Pfizer and Moderna vaccines was compiled from February 24th, 2020, to May 28th, 2021, using the Twint Python tool. Sentiment analysis was conducted using the Valence Aware Dictionary and Sentiment Reasoner (VADER) tool[2]. The stock price data was extracted from Yahoo Finance over the same period of time. Time lines of influencing events were constructed by identifying days for which a significant change in stock price or tweet sentiments occurred and then matching those days to significant events related to the pandemic. Our results suggest that news relating to vaccine clinical trials, vaccine distribution, the development of COVID-19 therapetiucs, government public health guide lines, government funding of vaccine development and the distribution and eligibility guidelines for the vaccines and PPE distribution are factors that likely influence change in the public’s positive, negative and neutral opinions on the vaccines. 
Furthermore, our analysis indicates that, to combat the spread of misinformation, public health units should devote more resources to ensuring that information about vaccine eligibility and vaccine distribution is accessible to the public.







"

What pandemic events have had the most influence on public sentiment towards the AstraZeneca, Pfizer and Moderna COVID-19 Vaccines?

Fake news is prevalent online due to the numerous social networks available on the internet. The COVID-19 pandemic started a new wave of health-related fake news that has increased the 
spread of misinformation. The tweets obtained from the ""CoAID"": COVID-19 Healthcare Misinformation Dataset contains healthcare information related to COVID-19, their reliable labels, and user engagement features. This paper applied machine learning (ML), deep learning (DL) and transformer based ML models (BERT) on the dataset. The dataset was first extracted using the computer software Hydrator, and included 252471 tweets from December 2019 to September 2020. The data was then preprocessed and vectorized. Exploratory analysis of the raw dataset was first done in Excel to find the correlation between the different tweet features. Classifying the tweets as ""real"" or ""fake"" was done in three ways: via Traditional Supervised Learning models, Traditional Natural Language Processing (NLP) models, and Deep Learning models. An algorithm was developed once the ideal NLP model was selected to further improve classification metrics. Traditional Supervised Learning models involved selecting variables about the tweets such as the number of retweets, and using those features for classification. Traditional NLP methods applied conventional models such as Naive Bayes (NB) and Random Forest (RF) models to the vectorized data. Finally, three deep learning models were created: a Deep Neural Network (DNN) model, a Recurrent Neural Network (RNN), and an RNN model that used a Bidirectional Encoder Representations from Transformers (BERT) model for word embedding. 
Overall, the best model to use was the RNN model since it performed well on the test dataset, and had the best classification metrics when exposed to data it had never seen before. 
An example of the deployed model our team created called ""TweetScan"" can be found here. It ultimately illustrates how the results of this paper can be applied to reduce the spread of 
misinformation, especially concerning recent issues such as vaccination acceptance.




Are You Spreading Misinformation? Analyzing How Influential Twitter Users Contribute to the Spread of COVID-19 Information

Downloads

Next from STEM Fellowship Inter-University Big Data Challenge 2021

The Plebeian Algorithm A Democratic Approach to Censorship and Moderation

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES