• Project number: F 2390
• Institution: Federal Institute for Occupational Safety and Health (BAuA) / N.N.
• Status: Completed Project

Description:

An important task of the market surveillance authorities is to identify dangerous products and to remove them from the European market. This applies to products sold in brick-and-mortar shops, as well as to products traded online. However, the large number of online retailers and products on sale, and complex distribution structures (online shops, platform providers, etc.) make efficient surveillance difficult.

The aim of this project was to examine whether automated analytical procedures can efficiently support the market surveillance authorities in controlling online trading. For this purpose, the project evaluated the use of different machine learning methods.

A software-based tool has been created to identify dangerous products from customer reviews. Therefore, data mining technologies were used to detect patterns in large sets of data automatically. In this project, special attention was paid to supervised learning. Methods based on weakly supervised learning, i.e. methods that require very small amounts of training data, proved to be particularly suitable. These methods made it possible to group potentially dangerous products systematically into risk categories (e.g. mechanical risk or fire risk) by analysing customer reviews. The products were then sorted and prioritised depending on the product type and risk assessment. In a next step, these product data were forwarded to market surveillance authorities for further inspection and assessment. In a field study, this approach made first achievements possible.

At the same time, the current legal framework limits automated monitoring of online trade, as the use of data mining technologies by market surveillance authorities remains unclear. Because of this, a legal expertise was commissioned. Its aim was to evaluate the legal possibilities for the automated analysis of risk assessments on online platforms. It became clear that the legal framework needs to be clarified in order to ensure that the authorities can act with sufficient legal certainty.

The intention is to continue developing the prototype software tool and the legal framework. Both is supposed to be done in collaboration with the market surveillance authorities. In future, the software tool is also to be made available to enforcement authorities for use in their own areas of jurisdiction.