Dataset Information

Influenza-like illness surveillance on Twitter through automated learning of naive language.

ABSTRACT: Twitter has the potential to be a timely and cost-effective source of data for syndromic surveillance. When speaking of an illness, Twitter users often report a combination of symptoms, rather than a suspected or final diagnosis, using naïve, everyday language. We developed a minimally trained algorithm that exploits the abundance of health-related web pages to identify all jargon expressions related to a specific technical term. We then translated an influenza case definition into a Boolean query, each symptom being described by a technical term and all related jargon expressions, as identified by the algorithm. Subsequently, we monitored all tweets that reported a combination of symptoms satisfying the case definition query. In order to geolocalize messages, we defined 3 localization strategies based on codes associated with each tweet. We found a high correlation coefficient between the trend of our influenza-positive tweets and ILI trends identified by US traditional surveillance systems.

SUBMITTER: Gesualdo F

PROVIDER: S-EPMC3853203 | biostudies-literature | 2013

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Influenza-like illness surveillance on Twitter through automated learning of naïve language.

Gesualdo Francesco F Stilo Giovanni G Agricola Eleonora E Gonfiantini Michaela V MV Pandolfi Elisabetta E Velardi Paola P Tozzi Alberto E AE

PloS one 20131204 12

Twitter has the potential to be a timely and cost-effective source of data for syndromic surveillance. When speaking of an illness, Twitter users often report a combination of symptoms, rather than a suspected or final diagnosis, using naïve, everyday language. We developed a minimally trained algorithm that exploits the abundance of health-related web pages to identify all jargon expressions related to a specific technical term. We then translated an influenza case definition into a Boolean que ...[more]

PMID: 24324799

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:BACKGROUND:Uptake of medicinal drugs (preventive or treatment) is among the approaches used to control disease outbreaks, and therefore, it is of vital importance to be aware of the counts or frequencies of most commonly used drugs and trending topics about these drugs from consumers for successful implementation of control measures. Traditional survey methods would have accomplished this study, but they are too costly in terms of resources needed, and they are subject to social desirability bias for topics discovery. Hence, there is a need to use alternative efficient means such as Twitter data and machine learning (ML) techniques. OBJECTIVE:Using Twitter data, the aim of the study was to (1) provide a methodological extension for efficiently extracting widely consumed drugs during seasonal influenza and (2) extract topics from the tweets of these drugs and to infer how the insights provided by these topics can enhance seasonal influenza surveillance. METHODS:From tweets collected during the 2012-13 flu season, we first identified tweets with mentions of drugs and then constructed an ML classifier using dependency words as features. The classifier was used to extract tweets that evidenced consumption of drugs, out of which we identified the mostly consumed drugs. Finally, we extracted trending topics from each of these widely used drugs' tweets using latent Dirichlet allocation (LDA). RESULTS:Our proposed classifier obtained an F1 score of 0.82, which significantly outperformed the two benchmark classifiers (ie, P<.001 with the lexicon-based and P=.048 with the 1-gram term frequency [TF]). The classifier extracted 40,428 tweets that evidenced consumption of drugs out of 50,828 tweets with mentions of drugs. The most widely consumed drugs were influenza virus vaccines that had around 76.95% (31,111/40,428) share of the total; other notable drugs were Theraflu, DayQuil, NyQuil, vitamins, acetaminophen, and oseltamivir. The topics of each of these drugs exhibited common themes or experiences from people who have consumed these drugs. Among these were the enabling and deterrent factors to influenza drugs uptake, which are keys to mitigating the severity of seasonal influenza outbreaks. CONCLUSIONS:The study results showed the feasibility of using tweets of widely consumed drugs to enhance seasonal influenza surveillance in lieu of the traditional or conventional surveillance approaches. Public health officials and other stakeholders can benefit from the findings of this study, especially in enhancing strategies for mitigating the severity of seasonal influenza outbreaks. The proposed methods can be extended to the outbreaks of other diseases.

Project description:BackgroundInternet-based surveillance systems to monitor influenza-like illness (ILI) have advantages over traditional (physician-based) reporting systems, as they can potentially monitor a wider range of cases (i.e. including those that do not seek care). However, the requirement for participants to have internet access and to actively participate calls into question the representativeness of the data. Such systems have been in place in a number of European countries over the last few years, and in July 2009 this was extended to the UK. Here we present results of this survey with the aim of assessing the reliability of the data, and to evaluate methods to correct for possible biases.MethodsInternet-based monitoring of ILI was launched near the peak of the first wave of the UK H1N1v influenza pandemic. We compared the recorded ILI incidence with physician-recorded incidence and an estimate of the true number of cases over the course of the epidemic. We also compared overall attack rates. The effect of using different ILI definitions and alternative denominator assumptions on incidence estimates was explored.ResultsThe crude incidence measured by the internet-based system appears to be influenced by individuals who participated only once in the survey and who appeared more likely to be ill. This distorted the overall incidence trend. Concentrating on individuals who reported more than once results in a time series of ILI incidence that matches the trend of case estimates reasonably closely, with a correlation of 0.713 (P-value: 0.0001, 95% CI: 0.435, 0.867). Indeed, the internet-based system appears to give a better estimate of the relative height of the two waves of the UK pandemic than the physician-recorded incidence. The overall attack rate is, however, higher than other estimates, at about 16% when compared with a model-based estimate of 6%.ConclusionInternet-based monitoring of ILI can capture the trends in case numbers if appropriate weighting is used to correct for differential response. The overall level of incidence is, however, difficult to measure. Internet-based systems may be a useful adjunct to existing ILI surveillance systems as they capture cases that do not necessarily contact health care. However, further research is required before they can be used to accurately assess the absolute level of incidence in the community.

Dataset Information

Influenza-like illness surveillance on Twitter through automated learning of naive language.

Publications

Influenza-like illness surveillance on Twitter through automated learning of naïve language.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure