AI-driven Big Data Analytics: Crisis Event Detection via Social Media Data

Blog - 2021-09-14

Pantelis Kyriakidis, Despoina Chatzakou, Ourania Theodosiadou, Theodora Tsikrika, Stefanos Vrochidis, Centre for Research and Technology Hellas (CERTH)

Social media have become one of the most preferable ways to share opinions and news, with around 53% of the world’s population using social media in 2021, a significant 13% increase compared to 2020 [1]. With this high impact, it is obvious that many social media platforms designed for news dissemination and opinions exchange (such as microblogging services like Twitter) often contain useful information that can be utilised for different purposes. One of those would be during emergency situations; when a crisis strikes, information gathered from platforms like Twitter could play a vital role in almost real-time incident detection, updating about potential consequences of a crisis event and its multilevel impact to the community.

To this end, there is a significant need to develop effective methods for highlighting and locating timely and valuable information about crisis events of interest. Focusing on Twitter (one of the most popular social media platforms worldwide [2]), the limited length of posts predisposes users to be more to the point and convey as much information as possible in a short text, which is often highly noisy and unstructured, as there is no norm for formal writing like in news sites. Therefore, due to these inherent difficulties, the analysis of Twitter data has proven to be a difficult task [3].   

In the context of the AI-driven big data analytics capabilities developed within INFINITY, the event detection framework aims to equip Law Enforcement Agencies (LEAs) with a tool capable of detecting crisis events of their interest (such as explosions, bombings, and shootings) in an effort to further improve their investigation capabilities. In particular, with a view to address the aforementioned issues that are often encountered in social media platforms like Twitter, state-of-the-art deep learning methods will be utilised and further improved within INFINITY to deliver high quality information to relevant stakeholders.

Figure 1 presents an example that highlights the importance of developing effective methods for the timely detection of crisis events. In this example, the green line depicts activity related to crisis events; in particular, it depicts the number of Twitter posts (y axis) in a specific timeframe (x axis) which were identified as related to a crisis event by the INFINITY event detection tool. The red line indicates tweets identified as not related to an event. Careful observation of the identified peaks can give an edge in early event detection that may deserve special attention.

Graph


                                                                              Figure 1. Online activity regarding event and non-event related posts.

In addition to analysing online activity as to whether it relates to crisis events in general, the INFINITY event detection tool is able to focus on more fine-grained analysis where specific events of LEAs interest are detected and highlighted. Figure 2 depicts an illustrative example where the focus is on three highly important crisis events related to the security of individuals in a society: bombings, shootings and explosions, the detection of which can further enhance the rapid detection and awareness of crisis events.

Graph


                                                                              Figure 2. Online activity regarding specific types of crisis events.

Finally, Figure 3 depicts some indicative examples of posts identified as related to a crisis event and in particular to an explosion. Focusing and examining only the posts of interest may prove to be quite effective in gathering valuable information that could further enhance the overall investigation.

Graph

                                                                 

                                                                           Figure 3. Examination of suspicious posts related to a crisis event of LEAs interest.

Overall, the INFINITY event detection framework allows for a rapid and effective identification of online textual content related to crisis events that are of LEAs interest. The already developed deep learning based method permits the early identification of crisis situations even with short and noisy texts (e.g. texts published on microblogging platforms such as on Twitter) and fast extraction of valuable insights. To this end, the INFINITY event detection framework has the potential to be a useful tool for LEAs towards providing assistance and support during an investigation.

Note: The dataset used for this research is the CrisisLexT26 [4] which is publicly available and has further been pseudonymised.

References

[1] Simon Kemp. 2021.  DIGITAL 2021: GLOBAL DIGITAL OVERVIEW. https://datareportal.com/reports/digital-2021-global-overview-report

[2] Statista. 2021. Twitter: monthly active users worldwide | Statista. https://www.statista.com/statistics/282087/number-of-monthly-active-twitter-users/

[3] Kourosh Alizadeh. 2021. Limitations of Twitter Data. https://towardsdatascience.com/limitations-of-twitter-data-94954850cacf

[4] Alexandra Olteanu, S. Vieweg, and C. Castillo. 2015. What to Expect When the Unexpected Happens: Social Media Communications Across Crises. Proceedings of the 18th ACM Conference on Computer Supported Cooperative Work Social Computing (2015). https://crisislex.org/data-collections.html#CrisisLexT26