Does this tweet contain self-reported COVID19 symptoms?Symptoms can be about the person tweeting or someone they are referring to directly. They should be current and not about past events. If you are unsure about your annotation, hit the “skip” button.”
April 6, 2020
@mention @mention Bonjour, Est-ce que vous savez si ce vendeur sur Amazon est fiable ? Il paraît que les arnaques se sont multipliées... car en effet, le port du masque est conseillé. Il aurait sûrement pu éviter des contaminations par les porteurs à symptomatiques ...
What is this about?
In this study, we aim to create a model of how tweets about self-reported COVID19 symptoms can help predict upcoming pandemic waves, and more generally the rise and fall of the disease. To that end, we crawled public tweets from the Paris region filtered by symptoms keywords, and plotted them in time (see the graph below).
However, this filtering is very crude, e.g people don't only tweet about symptoms when they are currently falling sick, but also about that one time a year ago when they fell sick, or when talking about the general news.
To filter out such false-positives we need your help! Which of these tweets are describing an acute symptom and which ones don't? Your contribution will make a direct impact!
If you want to learn more about the people behind this project you can visit our About page.
|Depuis Jeudi dernier je mange rarement j’ai des courbatures Migraine, toux aiguë||yes||July 26, 2022, 8:26 p.m.|
|J’ai genre tout les symptômes 😲 [url]||yes||July 26, 2022, 8:25 p.m.|
|Quand j’ai plus de goûter au cabinet 🥺 [url]||no||July 4, 2022, 4:37 p.m.|
|J’ai mal à la gorge||yes||July 2, 2022, 4:04 p.m.|
|@mention @mention et une agréable fièvre pour nous 😂||skip||June 20, 2022, 4:57 a.m.|
Do you want to run your own data analyses?
You can download all 8669 annotations (including a UUID referring to each annotator session) to create your own filtering or machine learning algorithm. Please visit our GitHub repository if you want to contribute!
Below we visualise the number of tweets containing COVID19 symptoms using 7 days averaging windows, both using our simple filtering system (orange), or the improved filtering using crowd-sourced annotation (blue). To build that curve, we only take into account tweets that have been labelled as "yes" at least 50% of the time ; we then show the daily rate of tweets labelled as "yes" multiplied by the daily number of tweets containing symptoms.
We show as a reference the number of passages to emergencies (see Analysis for more information).