Machine learning

Data-driven decision-making and machine learning have recently emerged as methods with high utility for many different applications. We integrate machine learning into our search application to provide user-specific search. Data such as browser type, operating system, type of device, time of day, time of year, or location allow conclusions to be drawn about the user and their needs. The geographic classification of customers is based on the Nielsen areas known from market research within Germany and, if relevant for the store, country-based outside Germany. It is therefore easy to imagine that the owner of an iPhone in Berlin at 10 a.m. and the owner of a Windows 7 PC in Lower Bavaria at 6 p.m. have completely different needs for the same search.

The best thing about machine learning, however, is that it does not use ideas or insider knowledge to categorize users, as this is purely derived from the collective behavior of other users. The prerequisite for this possibility is, of course, the collection of data that comes before the adaptation of search results. Only when enough statistically significant data has been collected and the values for the machine learning model have been converged, does it make sense to activate the user-specific search?

With this search, it is now possible to present visitors from different user groups with a better search experience tailored to them, and this in turn has an impact on the conversion rate.

The processing of the search results by machine learning is carried out by means of Rescoring. Thus, the content level of the search comes first, followed by the individual selection of suitable products for the user.


For machine learning to work, the feature must be turned on in the Makaira backend and the processing of user-specific data must be approved in the Connect module.