Please note: The algorithm descriptions in English have been automatically translated. Errors may have been introduced in this process. For the original descriptions, go to the Dutch version of the Algorithm Register.

Search application cold cases

This algorithm helps retrieve information in cold case files. It uses a language model to search for the meaning of words and not just the exact words.

Last change on 28th of January 2025, at 13:28 (CET) | Publication Standard 1.0
Publication category
Other algorithms
Impact assessment
DPIA, ...
Status
In use

General information

Theme

Public Order and Safety

Begin date

2024-09

Contact information

https://www.politie.nl/

Responsible use

Goal and impact

The algorithm helps detectives find information in a cold case file. Instead of searching for exact words (as with CTRL-F), it looks for the meaning of words. This makes it possible to find information faster and more completely.

Considerations

This is about doing justice to victims and the victim's relatives. The police want to continue investigating whether there are chances to solve a cold case. Files have gone through the hands of several detectives. This can make retrieving information difficult and time-consuming. With the cold cases search application, information can be found much faster.


Human intervention

Investigators start by reading the file to get a good understanding of the cold case. Sometimes they want to find information they have read before, or they look for new details. The cold cases search application helps with this by quickly finding pieces from the case file that are important. The search results show the words/phrases that have a semantic relationship with the search query. The investigator can therefore determine which search result is relevant and can read the original file from the search result. This information can help the searcher further in the investigation.

Risk management

A Data Protection Impact Assessment (DPIA) and an ethics quick scan were carried out. In addition, an Ethics & Law plan of action and assessment were prepared (based on the Ethical Guidelines for Trustworthy AI) and the General Court of Auditors' Algorithm Framework was checked. Measures were taken based on the risks and recommendations.

Legal basis

Article 8 Wpg when it comes to mutations in the context of regular police practice.


Article 9 Wpg when it comes to official reports drawn up from the initial investigation.


Links to legal bases

Wet politiegegevens: https://wetten.overheid.nl/jci1.3:c:BWBR0022463&paragraaf=2&artikel=8&z=2023-11-01&g=2023-11-01

Impact assessment

  • Data Protection Impact Assessment (DPIA)
  • Quickscan ethiek

Operations

Data

The algorithm uses an open-source language model that only supports Dutch. No police data was used in the development of this model.

Technical design

The user selects a cold case file to search in it. The user enters a query through the user interface, in the form of text. This text is converted into a vector representing the meaning of the sentence. It then calculates the 'distance' between this vector and the vectors of sentences or paragraphs from the case file. The result closest to the query entered is considered the most relevant.

The search results are presented in a list, ranked by relevance. For each result, the sentence or paragraph, the source, and the page number of the document are displayed. The words or phrases that, according to the algorithm, have a semantic relationship with the user's query are shaded yellow.

The search query is not stored in the Elasticsearch database. It may, however, be temporarily stored in the user interface so that the user can view previous searches.

External provider

Open source

Similar algorithm descriptions

  • This algorithm helps Customs to select goods for inspection based on risk. It uses declaration data from companies and considers whether or not there are risks of inaccuracies in the declarations.

    Last change on 28th of February 2024, at 12:29 (CET) | Publication Standard 1.0
    Publication category
    Impactful algorithms
    Impact assessment
    Field not filled in.
    Status
    In use
  • This algorithm helps the ACM find price announcements in messages that might say something about tacit or explicit alignment. This information by itself does not trigger action, but it can be a reason for inspectors to further investigate a possible infringement.

    Last change on 4th of October 2024, at 13:36 (CET) | Publication Standard 1.0
    Publication category
    Other algorithms
    Impact assessment
    Field not filled in.
    Status
    In use
  • This algorithm helps Customs to select goods for inspection based on risk. It uses declaration data from companies and considers whether or not there are increased risks of export and transit of strategic goods in customs declarations.

    Last change on 9th of December 2024, at 11:56 (CET) | Publication Standard 1.0
    Publication category
    Impactful algorithms
    Impact assessment
    Field not filled in.
    Status
    In use
  • This algorithm helps Customs to select goods for inspection based on risk. It uses declaration data from companies and considers whether or not there are risks of import, export and transit of torture equipment in customs declarations.

    Last change on 10th of December 2024, at 12:40 (CET) | Publication Standard 1.0
    Publication category
    Impactful algorithms
    Impact assessment
    Field not filled in.
    Status
    In use
  • The algorithm uses sensor data to make a calculation for predictions of slipperiness risks. These predictions are used to determine where to grit (preventively), in consultation with meteorologists. In addition, the algorithm can issue various alarms, for example when there is a chance of wet road sections freezing.

    Last change on 10th of July 2024, at 8:18 (CET) | Publication Standard 1.0
    Publication category
    Other algorithms
    Impact assessment
    Field not filled in.
    Status
    In use