Turning unstructured food order data into valuable information

Lieferando.de, one of Germany’s main food delivery platform, lists more than 30.000 restaurants and processes millions of orders per month. Nonetheless, its ability to analyse food order data for business intelligence used to be limited due to the unstructured nature of that information. idalab developed a text classification algorithm which enables precise identification of meal types from text data. This forms the backbone of Lieferando.de’s comprehensive data science initiative, improving business analytics and facilitating highly-customised recommendations.

 

Situation

  • Lieferando.de is a market place for food orders
  • Low integration hurdle for partner restaurants essential for fast expansion
  • Menu items formerly submitted as unstructured text without any category information or other metadata

Approach

  • Development of a customised natural language processing algorithm for hierarchical classification (>1000 categories)
  • Calibration based on a small set of hand-labelled texts

Impact

  • High precision (>90%) plus reliable confidence measure to manually inspect difficult cases
  • Use cases: improved search, user profiling by tastes, dramatically better business intelligence capabilities

Back to cases overview