The way we eat and think about food is a part of our culture and a product of our time. Today, you can find even the most obscure recipe simply by going online. Before the internet, newspapers and magazines were an important source of culinary inspiration, and by going back in time these can provide valuable insights into the evolution of food customs.
In this paper, researchers from Amsterdam extract and structure recipes found in Dutch newspapers published between 1945 and 1995. Relying on digitized and OCR’ed (Optical Character Recognition) text from the National Library of the Netherlands, they use machine learning methods to construct a historical database of more than 27,000 recipes. To enrich the data, their workflow separates and tags quantities, units and ingredients. Using GBIF-mediated data, they even match naturally sourced ingredients against scientific names and determine their origin.
Combining language processing, machine learning and semantic structuring, this work provides a valuable source of data on food culture. One example of a changing concept is how some vegetarian recipes often contained animal products (e.g. chicken broth and fermented shrimp). The authors, however, leave the actual interpretation of the results to humanities researchers, as the code and resulting database are available for anyone to use, browse and analyze.