In collaboration with Amsterdam-Amstelland police (the Netherlands): the largest regional police force in the Netherlands and one of the first in the world to adopt the Intelligence Led Policing management paradigm.
Domestic violence prevention: In this project we annually analyzed around 20000 statements made by victims of a violent incident to the police using text mining, formal concept analysis and emergent self-organizing maps. In this research project, we were able to expose multiple unknown niche cases, faulty case labelings, early warning indicators, etc. which led to a redefinition of the concept of domestic violence employed by the responsible unit as well as a significant improvement of police training on the subject. Another result of our work was the development of an automated case labeling system which automatically recognizes incoming suspicious cases. Operationalization of the system led to significant time and cost savings and a higher recognition rate, while at the same time false positives were reduced. Our approach outperformed all previous approaches using traditional data mining methods and was the first to make it into operational policing practice.
Identification of human trafficking suspects and victims: In this project we aimed at developing a semi-automated text analysis system which could be used to identify unknown suspects and victims of human trafficking. We initially analyzed 266157 unstructured suspicious activity reports combining text mining and data discovery techniques. Multiple unknown suspects and victims were found, and their social network was further analyzed with our system. Special investigation techniques employed by the police as a result of these findings, resulted in multiple gangs being imprisoned. The system is currently used with success on a daily basis by the anti-human trafficking team for identifying in real time potential suspects and victims from big dynamic text collections.
Terrorism prevention: In the aftermath of the terrorist attacks which happened in Europe since 2004, multiple law enforcement and intelligence services in the Netherlands developed a radicalization model which describes the expected process a potential Jihadist goes through before committing an attack. We operationalized this model to make it applicable for automatically detecting threats from large quantities of unstructured text. Our system is currently being used for automatically scanning intercepted email traffic, suspicious activity reports, etc. to rapidly identify potential threats to national security.
Automated recognition and analysis of pedophile chat conversations: The goal of this project was to develop a methodology for quickly gaining insight in and flagging escalating chat conversations. The system is currently in use to automatically search through chat conversations found on confiscated computers. Its adoption resulted in significant time and cost savings.
In collaboration with GZA Hospitals, Antwerp (Belgium): a regional hospital group with over 1000 patients beds and one of the first in Belgium to integrate its care with the IT systems of the hospital.
Leveraging process and data discovery for improving quality and efficiency of care: In this project we analyzed 1602421 medical records containing information on care activities performed to patients for whom care was organized by the methodology of clinical pathways. GZA hospitals implemented 42 care pathways which are fully logged into the hospital’s electronic patient record system. Due to this large amount of data, it becomes impossible to monitor quality of care and patient safety in a manual way. We used the C++ language and formal concept analysis for preparing and analyzing the data. Domain experts were able to identify deviations from the prescribed clinical pathway, root causes for these deviations and exceptional care trajectories, value leaks leading to suboptimal care and financial losses. The directory board and the different multidisciplinary teams used the results for improving the quality and efficiency of care and patient safety.
Evaluating patient experience through text mining and text discovery: In this project we analyzed 40539 unstructured textual statements made by patients, written on their day of discharge, about their experience of care during their hospitalization. We used the C++ language and formal concept analysis for the analyses. We implemented state of the art techniques from the areas of text mining and algebra for performing these analyses in an effective, thorough and rapid manner. Results of the analysis have been used by the directory board for identifying unknown problems and novel solutions to improve patient experience.
Identifying patient groups for which the hospital needs to open a novel unit to sustain its quality of care delivery and reduce bottlenecks at discharge: In this project we linked together and analyzed heterogeneous data repositories, from different hospital departments, containing information on discharge destinations of 3148 patients, optimal and actual lengths of stay, clinical patient characteristics such as diagnosis, severity of illness, etc. We used the C++ language and formal concept analysis for preparing and analyzing the data. Domain experts were able to identify several patient groups in a very short time span, for which there was a need for opening a novel unit. Our systems were also used for rapidly identifying the characteristics of patients which led to suboptimal care. The results were used to request funding from the Belgian Government for opening a novel unit and permission was received in March 2014.