Geographic variability of Twitter usage characteristics during disaster events

Figures & data

Table 1. Set of Keywords used to query Twitter Streaming API.

List of keywords used for the querying
Tsunami	flood	Earthquake
Landslide	earth quake	fore shock
fore-shock	after shock	after-shock
landslide	land slide	Avalanche
rockfall	rock fall	mud slide
mudslide	earth slip	earth-slip
cloudburst	cloud burst	heavy rainfall
extensive rainfall	heavy rain	extensive rain
rain storm	forest fire	inundation
overflow	flash-flood	−

Table 2. Data-set details.

Size	Tweets	Start time	End time
2.54 GB	488,175	Wednesday 24 August 2016 08:57	Thursday 25 August 2016 08:57

Table 3. User-based attributes.

User-based features	Description
Registration age	The time passed since the author registered their account
Statuses count	The number of Tweets sent by the user
Followers count	Number of people following this user
Friends count	Number of people user is following
Verified	If the account has been verified
Has description	A non-empty bio
Has URL	A non-empty homepage URL

Figure 1. Myanmar earthquake geographic feature granularity according to Geonames.

Figure 2. Italy earthquake geographic feature granularity according to Geonames.

Figure 3. Occurrence of toponyms over time in Italy.

Figure 4. Occurrence of toponyms over time in Myanmar.

Table 4. Differences between credibility-related attributes.

Attribute	Italy	Myanmar
Friends count	1320 ± 3839	1073 ± 2369
Statuses count	31,498 ± 79,831	52,067 ± 101,762
Followers count	2082 ± 5479	3966 ± 22,622
Account age	5.32 ± 2.25	3.45 ± 2.53

Table 5. Confusion matrix for Italy (Case 1).

Actual class
Predicted class	Class	Information	Not information	Precision
	Information	147	3	98%
	Not information	11	139	92.667%
	Recall	93.038%	97.887%	−

Table 6. Confusion matrix for Myanmar (Case 1).

Actual class
Predicted class	Class	Information	Not information	Precision
	Information	133	17	88.667%
	Not information	12	138	92%
	Recall	91.724%	89.032%	−

Table 7. Confusion matrix for Myanmar (Case 2).

Actual class
Predicted class	Class	Information	Not information	Precision
	Information	149	1	99.333%
	Not information	11	139	92.667%
	Recall	93.125%	99.286%	−

Table 8. Confusion matrix for Italy (Case 2).

Actual class
Predicted class	Class	Information	Not information	Precision
	Information	141	9	94%
	Not information	37	113	75.333%
	Recall	79.213%	92.623%	−

Table 9. Confusion matrix for Italy (Case 3).

Actual class
Predicted class	Class	Information	Not information	Precision
	Information	146	4	97.333%
	Not information	36	114	76%
	Recall	80.22%	96.61%	−

Table 10. Confusion matrix for Myanmar (Case 3).

Actual class
Predicted class	Class	Information	Not information	Precision
	Information	146	4	97.333%
	Not information	37	113	75.333%
	Recall	79.781%	96.581%	−