Dataset Information

Public Discussion of Anthrax on Twitter: Using Machine Learning to Identify Relevant Topics and Events.

ABSTRACT:

Background

Social media allows researchers to study opinions and reactions to events in real time. One area needing more study is anthrax-related events. A computational framework that utilizes machine learning techniques was created to collect tweets discussing anthrax, further categorize them as relevant by the month of data collection, and detect discussions on anthrax-related events.

Objective

The objective of this study was to detect discussions on anthrax-related events and to determine the relevance of the tweets and topics of discussion over 12 months of data collection.

Methods

This is an infoveillance study, using tweets in English containing the keyword "Anthrax" and "Bacillus anthracis", collected from September 25, 2017, through August 15, 2018. Machine learning techniques were used to determine what people were tweeting about anthrax. Data over time was plotted to determine whether an event was detected (a 3-fold spike in tweets). A machine learning classifier was created to categorize tweets by relevance to anthrax. Relevant tweets by month were examined using a topic modeling approach to determine the topics of discussion over time and how these events influence that discussion.

Results

Over the 12 months of data collection, a total of 204,008 tweets were collected. Logistic regression analysis revealed the best performance for relevance (precision=0.81; recall=0.81; F₁-score=0.80). In total, 26 topics were associated with anthrax-related events, tweets that were highly retweeted, natural outbreaks, and news stories.

Conclusions

This study shows that tweets related to anthrax can be collected and analyzed over time to determine what people are discussing and to detect key anthrax-related events. Future studies are required to focus only on opinion tweets, use the methodology to study other terrorism events, or to monitor for terrorism threats.

SUBMITTER: Miller M

PROVIDER: S-EPMC8277308 | biostudies-literature |

REPOSITORIES: biostudies-literature

ACCESS DATA

Similar Datasets

Project description:BackgroundTwitter presents a valuable and relevant social media platform to study the prevalence of information and sentiment on vaping that may be useful for public health surveillance. Machine learning classifiers that identify vaping-relevant tweets and characterize sentiments in them can underpin a Twitter-based vaping surveillance system. Compared with traditional machine learning classifiers that are reliant on annotations that are expensive to obtain, deep learning classifiers offer the advantage of requiring fewer annotated tweets by leveraging the large numbers of readily available unannotated tweets.ObjectiveThis study aims to derive and evaluate traditional and deep learning classifiers that can identify tweets relevant to vaping, tweets of a commercial nature, and tweets with provape sentiments.MethodsWe continuously collected tweets that matched vaping-related keywords over 2 months from August 2018 to October 2018. From this data set of tweets, a set of 4000 tweets was selected, and each tweet was manually annotated for relevance (vape relevant or not), commercial nature (commercial or not), and sentiment (provape or not). Using the annotated data, we derived traditional classifiers that included logistic regression, random forest, linear support vector machine, and multinomial naive Bayes. In addition, using the annotated data set and a larger unannotated data set of tweets, we derived deep learning classifiers that included a convolutional neural network (CNN), long short-term memory (LSTM) network, LSTM-CNN network, and bidirectional LSTM (BiLSTM) network. The unannotated tweet data were used to derive word vectors that deep learning classifiers can leverage to improve performance.ResultsLSTM-CNN performed the best with the highest area under the receiver operating characteristic curve (AUC) of 0.96 (95% CI 0.93-0.98) for relevance, all deep learning classifiers including LSTM-CNN performed better than the traditional classifiers with an AUC of 0.99 (95% CI 0.98-0.99) for distinguishing commercial from noncommercial tweets, and BiLSTM performed the best with an AUC of 0.83 (95% CI 0.78-0.89) for provape sentiment. Overall, LSTM-CNN performed the best across all 3 classification tasks.ConclusionsWe derived and evaluated traditional machine learning and deep learning classifiers to identify vaping-related relevant, commercial, and provape tweets. Overall, deep learning classifiers such as LSTM-CNN had superior performance and had the added advantage of requiring no preprocessing. The performance of these classifiers supports the development of a vaping surveillance system.

Project description:ContextHousing is more than a physical structure-it has a profound impact on health. Enforcing housing codes is a primary strategy for breaking the link between poor housing and poor health.ObjectiveThe objective of this study was to determine whether machine learning algorithms can identify properties with housing code violations at a higher rate than inspector-informed prioritization. We also show how city data can be used to describe the prevalence and location of housing-related health risks, which can inform public health policy and programs.SettingThis study took place in Chelsea, Massachusetts, a demographically diverse, densely populated, low-income city near Boston.DesignUsing data from 1611 proactively inspected properties, representative of the city's housing stock, we developed machine learning models to predict the probability that a given property would have (1) any housing code violation, (2) a set of high-risk health violations, and (3) a specific violation with a high risk to health and safety (overcrowding). We generated predicted probabilities of each outcome for all residential properties in the city (N = 5989).ResultsHousing code violations were present in 54% of inspected properties, 85% of which were classified as high-risk health violations. We predict that if the city were to use integrated city data and machine learning to identify at-risk properties, it could achieve a 1.8-fold increase in the number of inspections that identify code violations as compared with current practices.ConclusionGiven the strong connection between housing and health, reducing public health risk at more properties-without the need for additional inspection resources-represents an opportunity for significant public health gains. Integrated city data and machine learning can be used to describe the prevalence and location of housing-related health problems and make housing code enforcement more efficient, effective, and equitable in responding to public health threats.