Rick Adderley: Abstract

Criminals rarely work on their own as they tend to commit crime from within an organic network of offenders, persons who may join and leave the network at any time. These criminals leave a data trail spanning multiple Police systems that are often not linked meaning that the same person may in present in several disparate data sets. When a criminal is arrested for a crime and whose details are entered into the relevant computer system, a unique reference number is generated. When that person is arrested again for another crime, the inputter should use a “look up” to determine any previous arrests and if already present in the system allocate the same URN. Due to pressures of work and lack of training occasionally this process does not work and the offender is allocated a new URN. A further complication is when the name is spelled slightly differently and/or the date of birth also differs slightly meaning that the inputter may regard this person as a new entry and therefore not link to the original entry. Most Police networking packages create criminal networks either based on a URN or combination of names. If there are discrepancies in either of these, the generated network will treat each different entry as the start point for a new network meaning that the actual network will not be complete.

In order to reduce crime and arrest offenders it is important that Law Enforcement personnel understand the nature of the individuals and how they interact within their network. It is not uncommon for an acquisitive crime network at two degrees of freedom to exceed 250 different people making it very difficult for an analyst to determine which offender(s) to target in order to disrupt that network’s activities.

This issue is further compounded when the criminal’s data is in one or more different data bases such as Custody, Intelligence and Stop & Search when they are not linked by URN. How do you know whether it is the same person?

This problem can be reduced by a process called Entity Resolution (ER) the aim of which is to select and match entities that have a high probability of relating to the same thing or person. ER considers an entity to be a database record or spreadsheet row and attempts to find duplicate or nearly identical records. Once resolved the entities (criminals) can be more efficiently networked and individuals modelled. The modelling reveals and prioritises those individuals who need to be targeted in order to maximise disruption with minimal effort.

Organised by BCS SGAI The Specialist Group on Artificial Intelligence http://www.bcs-sgai.org