Cases
Kaemingk: Feature Extraction
- 20,000 items
- Feature Extraction
- 90% automatically extracted
Kaemingk is a prominent distributor and importer of seasonal decorative goods. With a workforce of approximately 450 employees, Kaemingk each year showcases the most creative home decor collections for occasions such as Christmas, spring/summer, Easter, and Valentine’s Day.
Their product lineup features over 20,000 decorative items for both home and garden, which are exhibited in showrooms located in the Netherlands, Belgium, Germany, and the United States. They supply thousands of professional clients across more than 80 countries.
Challenge
In updating their IT application landscape, Kaemingk rolled out a new PIM system that necessitated a structured entry of all products in their catalog. The pre-existing product data was contained within various written descriptions that needed to be transformed into individual product attributes. Undertaking this transformation manually was highly time-consuming and required the attention of employees who were already engaged in other critical processes.
Solution
By leveraging advanced algorithms, the extraction of product attributes has been largely automated. This automation significantly decreases the need for manual data conversion, resulting in a uniform dataset.
Before automation, part of the conversion had been manually performed by product specialists. This initial data was used to train algorithms to identify which keywords in the text corresponded to each specific attribute. These keywords were further refined with input from product specialists to enhance the software’s ability to recognize accurate conversions.
Result
The conversion of the datasets for Christmas 2019 and Spring 2020 led to approximately 90% of the identified feature values being deemed confident enough for immediate approval. To ensure comprehensive coverage, a web application was developed that allowed product specialists to review and finalize products within their expertise. Both the application and the algorithms were created in collaboration with these specialists to guarantee quality and user-friendliness.
