Stage2020-INSA-Handicap-International/covid19-nowcast

Covid-19 Nowcast in Africa # covid19-nowcast-africa Detect and analyze the spread of COVID-19 in Africa thanks to social networks led by BETTINGER Matthieu, HAFID Salim and SADER Bruno for Humanity & Inclusion in collaboration with the Institut Gaston Berger. # Summary # Research phase ## Corpus and crawling Texts and documents forming the corpus are gathered from posts on social media, as well as articles from news websites for the usecase of category classification training. As of the time of writing, requests can be made to Twitter's Standard and Premium APIs (7 days, 30 days and full archive requests). Because of external complications, Facebook's GraphAPI could not be tested and therefore isn't handled yet. Category classification training has been done on articles with different manners of speech, available on news websites like the New York Times, Associated Press, BuzzFeedNews, and local news like HindustanTimes and KenyanNews. ## Volumetric Analysis The idea was to provide the user with a dynamic graph featuring multiple metrics that help better detect and analyze the spread of COVID-19 in HI's countries of intervention. By displaying multiple metrics on a single graph, one can visually see the correlations between metrics and infere useful information. See Web App Development > Frontend > Graphs for more details on the displayed metrics and the resulting feature. ## Topic Analysis The social media data being twitter data, it usually consists of a high number of tweets (i.e very short texts) that tackle the COVID-19 issue from a wide variety of angles. To ensure that the user gets a clear grasp of the different topics being talked about when talking about COVID-19 in a specific country, the tool provides a Topic Classification feature. See Web App Development > Frontend > Topics for more details on the resulting feature. ## Sentimental Analysis ### English For the english sentiment analyser, we decided to use a XLNet (XLNet: Generalized Autoregressive Pretraining for Language Unde …