Detection and monitoring algorithm Digital Economy

Consumer and Market Authority

This algorithm can give the ACM insight into trends, developments and potential problems that exist in the digital market. This information in itself does not trigger action, but it can be a reason for inspectors to further investigate a potential infringement.

Last change on 4th of October 2024, at 13:35 (CET) | Publication Standard 1.0

• Publication category: Other algorithms
• Impact assessment: Field not filled in.
• Status: In use

General information

Theme

Economy

Begin date

2024-07

Contact information

[e-mail]

Responsible use

Goal and impact

The markets overseen by the ACM are becoming increasingly digital. To properly supervise these digital markets, the Detection and Monitoring Algorithm Digital Economy has been developed. This algorithm gives a good picture of the current issues in the digital market. If further research shows that it is necessary, the ACM can intervene. This can be done by providing additional explanation, issuing a warning, a binding instruction, imposing an order under penalty, or imposing a fine. With this algorithm, the ACM contributes to a healthy economy by making markets work well for all people and businesses, now and in the future.

Considerations

The number of reports from businesses and consumers on the digital market that the ACM receives is low. So the ACM cannot rely on these reports alone if it wants to properly monitor whether companies are complying with the rules. The ACM sees that reports about the digital market are made on many websites. These reports can help identify trends, developments and problems in the digital market.

Human intervention

The information provided by this algorithm is assessed by inspectors and further investigated if necessary. So there is always human intervention because someone checks the information. The information provided by the algorithm does not in itself lead to action being taken, but it may be a reason for the ACM to further investigate a possible breach of the law.

Risk management

The information provided by this algorithm is one of the many sources the ACM uses to monitor. Inspectors always check that the information is relevant and reliable. The results of the algorithm are monitored and, based on the results and new insights, the algorithm can be adjusted.

Operations

Data

Online resources

Technical design

This tool actually comprises two algorithms. To make data insightful, it uses a tagging algorithm. This algorithm analyses and categorises the collected information by tagging it based on predefined search terms. These search terms are selected by ACM inspectors as relevant for ACM supervision. The tagging algorithm searches the dataset and identifies each message that contains a relevant search term. This allows inspectors to quickly gain insight into the parts of the data of interest to them. After the data is tagged, the information is presented in a dashboard. On this dashboard, messages containing relevant tags are displayed by tag. Inspectors can further filter these tags based on date, source, or by entering specific search terms. This dashboard provides a clear and user-friendly view of the tagged data, giving inspectors easy access to the information that is relevant to them.

It also uses a topic modelling algorithm to identify overarching topics within a collection of posts. Topic modelling is an algorithm used to identify abstract topics within a collection of documents. The process starts by converting tags into so-called "sentence embeddings", which is a series of numbers representing the meaning of the tags. It then looks at how best to group these sets of numbers, placing the messages that are closest to each other in terms of numbers into logical groups and clustering them into topics. The model knows how to self-discover topics within a collection of texts and can self-identify themes. This allows exploratory identification of topics and possible market issues. The algorithm uses a large language model to self-discover topics and determine trends based on the available information. For each forum, an algorithm assigns topics to pieces of text. The topic modelling process involves two main steps: using the underlying model and clustering the documents. The more data is available, the better the clusters can be formed and the more accurately topics can be identified.