Fig 1.
A framework of processes for identifying potential witnesses of events from evidence in micro-blogs.
Table 1.
Summary definitions of WA and IA [3, p.6].
Table 2.
Example text content to describe the evidence inference categories NE, OTG and NOTG [6, p.3].
Fig 2.
Illustrative image content to describe the evidence inference categories A) NE, B) OTG, and C) NOTG.
A) and C) are printed under a CC BY license, with permission from Marie Truelove, original copyright 2017. B) printed under a CC BY license, with permission from Rachael Hopkins, original copyright 2017.
Table 3.
A summary of evidence types for on-topic micro-blogs.
Fig 3.
Illustrative example of on-topic micro-blogs posted by a single micro-blogger that distinguish between on-hash and off-hash categories for a target event #AFLDonsPies.
The image and text content in this figure are similar to content posted by an example micro-blogger, and are for illustrative purposes. The images in the figure are printed under a CC BY license, with permission from Rachael Hopkins, original copyright 2017.
Table 4.
Examples of potential NOTG evidence from the off-topic text content.
Fig 4.
The sets representing evidence and micro-blogs posted by an example micro-blogger.
The micro-blogger MB1 has posted three micro-blogs M1–3 that include three text evidence T = {e1, e4, e6}, two images I = {e2, e5}, and two geotags G = {e3, e7}. The image and text content in this figure are similar to content posted by an example micro-blogger, and are for illustrative purposes. The images in the figure are printed under a CC BY license, with permission from Rachael Hopkins, original copyright 2017.
Fig 5.
The subsets of evidence defined for testing the example micro-blogger.
Five pieces of evidence e2, 3, 5–7 categorized as OTG or NOTG remain for inference testing within three micro-blogs MT1–3 after content categorized as NE is removed. The image and text content in this figure are similar to content posted by an example micro-blogger, and are for illustrative purposes. The images in the figure are printed under a CC BY license, with permission from Rachael Hopkins, original copyright 2017.
Fig 6.
The derivation of the combined inference category for each micro-blog posted by the example micro-blogger.
The micro-blogs MT1 and MT2 are categorized OTGM because all evidence contained are categorized OTG. MT3 is categorized MIXW because it contains evidence that are both OTG and NOTG. The image and text content in this figure are similar to content posted by an example micro-blogger, created for illustrative purposes. The images in the figure are printed under a CC BY license, with permission from Rachael Hopkins, original copyright 2017.
Fig 7.
The derivation of the test within result for each micro-blog posted by the example micro-blogger.
The test within result for MT1 is corroboration CORW, and the result for MT3 is conflict CONW. MT2 must be assigned no test within NTW. The image and text content in this figure are similar to content posted by an example micro-blogger, and are for illustrative purposes. The images in the figure are printed under a CC BY license, with permission from Rachael Hopkins, original copyright 2017.
Fig 8.
The derivation of the summary inference category for the example micro-blogger.
The micro-blogger MBT1 is assigned a summary inference category of MIXB as they have posted micro-blogs of mixed inference categories. The image and text content in this figure are similar to content posted by an example micro-blogger, and are for illustrative purposes. The images in the figure are printed under a CC BY license, with permission from Rachael Hopkins, original copyright 2017.
Fig 9.
The derivation of the test between category for the example micro-blogger.
Micro-blogger MBT1 is assigned the test between category corroboration CORB as the test set MBTT1 contains more than one micro-blog of the same combined inference category OTGM. The image and text content in this figure are similar to content posted by an example micro-blogger, created for illustrative purposes. The images in the figure are printed under a CC BY license, with permission from Rachael Hopkins, original copyright 2017.
Table 5.
A summary of the case study events [10].
Table 6.
A summary of event milestones with corresponding approximate time intervals.
Table 7.
A summary of the temporal filters for datasets by content sources.
Table 8.
An example of manually assigned mf for evidence by content source and inference category.
Table 9.
The number of content source for the ADon and ADoff datasets at the completion of the filtering processes.
Table 10.
The number of evidence by inference category and content source for the ADon and ADoff datasets.
Table 11.
The number of micro-blogs by summary inference category CM and corroboration for the ADon and ADcomb datasets.
Table 12.
The number of micro-bloggers by summary inference category CB and corroboration for datasets ADon and ADcomb.
Table 13.
The summary inference category CB, corroboration COR, and Bel(OTG) value and corresponding rank, for the example micro-bloggers by the ADon and ADcomb datasets.
Fig 10.
Micro-blogs and evidence identified for Sensor151.
The conflicting evidence e15 is identified at the end of the timeline. The image and text content are similar to that posted by the micro-blogger from the case study, and are for illustrative purposes. The images in the figure are printed under a CC BY license, with permission from Rachael Hopkins, original copyright 2017.
Fig 11.
Micro-blogs and evidence identified for Sensor6.
The conflicting evidence e3 is identified at the beginning of the timeline. The image and text content are similar to that posted by the micro-blogger from the case study, and are for illustrative purposes. The images in the figure are printed under a CC BY license, with permission from Rachael Hopkins, original copyright 2017.
Table 14.
Example miss-classified text by human annotators.
Table 15.
A summary of text and image classification results.
Table 16.
The number of micro-blogs by combined inference category CM and corroboration for the ADon_a and ADcomb_a datasets.
Table 17.
The number of micro-bloggers by summary inference category CB and corroboration category for the ADon_a and ADcomb_a datasets.
Table 18.
The summary inference category CB, corroboration COR, and Bel(OTG) value and corresponding rank, for example micro-bloggers by the ADon_a and ADcomb_a datasets.
Fig 12.
A comparison of approaches for deriving mf for geotag evidence.
w is computed using a Tri-cube Kernel function Eq (6), and mf(OTG) is a manual assignment based on a decision boundary for categorization of OTG or NOTG.