Data Science & Machine Learning

Smart Policing Hackathon: AI-driven crime investigation

From 9 – 11 December 2020, The Belgian Federal Police, VIAS and EY organized the first Smart Policing Hackathon. Arinti was selected as the winner of the Smart Investigation challenge, with their Cognitive Search & Graph Analytics solution Tafuta.

Ensuring security and quality of life in society is a constant challenge for the Police. Society and crime are constantly evolving, so the Police must also continue to innovate smartly. Remaining alert and agile to efficiently combat small, large and disruptive crime is essential for the Police.  

The Smart Policing Hackathon is a launchpad to boost smart innovation in the Police. How can they put innovation at the service of policing for citizens? And how can the Police as an organization innovate smartly in order to position itself as an attractive employer in the labor market? 

During 2 days, teams of start-ups, scaleups, corporates and students had the opportunity to collaborate and create digital and innovative solutions relevant for the Police. A diverse and dedicated team of experts and coaches assisted during the process. After the hacking, a mixed jury of high-level experts (First Chief Superintendent General Commissioner of the Federal Police – Marc De Mesmaeker, First Chief Superintendent, President of the Permanent Commission for the Local Police, Head of Police of the Local Police zone Westkust – Nicholas Paelinck, Inetum Corporate Vice President Benelux, Digital Champion Belgium, EU Governing Board Digital Skills and Jobs – Saskia Van Uffelen, Professor of sociology at the University of Liège, Dean of the Faculty of Social Sciences – Frédéric Schoenaers) selected the winners.  

Smart Investigation

The challenge we picked was all about finding innovative investigation techniques to improve the fight against serious and organized crime. There are many opportunities in the field of technology that could provide better support for current investigation techniques. The aim is to find out what these might be and what their potential applicability might be. We decided to focus on knowledge discovery and connecting the multitude of information available.

Participating in the hackathon allowed us to dive deep into the current way of working when it comes to investigations. We learned that the huge amount of information that is being created every day is hardly being used effectively. Entities like persons, vehicles, objects, telephone & bank account numbers, known affiliations to organizations and groups, and so on, have to be manually tagged in PV’s, reports and investigations, which is very time consuming. Analysts working on finding connections and links between these entities are extremely scarce within the Police, and there is no proactive way of information sharing, by for example “tracking” if there’s new information available about certain entities. Even when these entities are labeled correctly, there’s no concept- or synonym aware search available. On top of that, there’s a huge potential for information discovery in unstructured data like social media data, phone tab transcripts, and so on. In conclusion, we can definitely say that data is siloed and the potential of the data the Police are sitting on has hardly been explored.

Advanced Cognitive Search

We started the hackathon with a concrete scenario in mind: imagine that a body is found. The victim turns out to be a known drug-dealer from a large drug investigation a while ago. The inspectors working the case now have to find any known links to the victim as fast as possible, to catch the people responsible. There’s two crucial aspects in this research; The inspectors have to find any known cases where the victim’s name appears, and they have to find other people or entities connected to this case and individual.

We set out to prototype a Police-version of Tafuta, our Cognitive Search & Graph Analytics tool. We wanted to create a product that allowed investigators to query the huge amount of available criminal data and discover links in that data as fast as possible, in an visual and interactive way.

With Tafuta, agents and investigators will be able to exploit the full potential of the data, in a google (or Bing)-like experience. The tool understands documents, it interprets text and turns raw information into searchable content. It recognizes entities automatically and is concept aware, meaning it understands all the variations used to label a certain entity. We’ve also added a speech-to-text feature, so we can automatically transcribe audio files and search through for example wire taps. With the object detection feature we’ve added, we’re able to annotate graphic content and textually describe what a certain image shows, making that part of the searchable content as well. Investigators can also save their searches, to get proactive alerts when new information appears about an entity they are interested in. But that’s not where it ends. Investigating is connecting the dots. That’s why the Graph Analytics part is so important. Thanks to graph analytics we can build strong visual representations of how entities are connected and linked to each other. This will allow investigators to literally see how the dots are connected.

The value Tafuta will bring the Police is an improved distribution and spread of information. It will be easier to understand the links between criminal facts. Investigators and inspectors will have to spend less time on humdrum work like labelling the entities in PV’s and other documents. We believe not only the police will benefit from Tafuta, but also the juridical system in the broader sense. Better spread of information will lead to faster and more accurate research work, hopefully leading to lower rate of criminal offences.

Thanks a lot for the opportunity to participate in this hackathon. We’re looking forward to the further collaboration!

You can find additional blog-posts, news articles and reports about the #SmartPolicingHackathon here:

Newsletter

Stay tuned.