Fast automatic estimation of spatial Mercalli intensity based on social media

This dataset consists of metadata related to quakes recorded in Chile from January 2016 to June 2017. The dataset is a cross match between tweets and seimological records provided by the National Seismological Center. We collected 825310 tweets posted by 309749 different users for the time period indicated above. We matched each earthquake to its corresponding municipality data to create municipality-level data units. Data units considered tweets from the time at which an earthquake occurred until 30 minutes afterwards.

The dataset is provided into four .tsv/.csv files:

- municipalities.tsv consists of the description of the municipalities of our dataset, consisting of four columns separated by tabs. The columns corresponds to the municipality ID, the region ID, the name of the municipality and its population.

- municipality_location.tsv consists of the geodesic location of the municipality according to the coordinates of its capital. It has three columns separated by tabs. The first corresponds to the municipality ID, the second and third correspond to latitude and longitude, respectively.

- training.csv corresponds to data instances that describe cross matches between earthquakes and municipalities. Each row of the file describe a data instance with 15 columns separated by commas. The first one is the quake ID, the second one corresponds to the municipality ID. The following 12 columns correspond to features retrieved from the Twitter stream (e.g. number of tweets, etc). The last column indicate the intensity in the Mercalli scale.

- testing.csv correspond to data instances reserved for testing purposes. This file has the same organization considered for training.tsv