Automatic Extraction of Nutrition Information from Nutrition Strings

ACL ARR 2025 May Submission2852 Authors

Abstract:

Finding the right food in a supermarket for someone's dietary needs is challenging due to the large variety of food products.
One possible solution is to build a nutrition information dataset and a food search engine. This engine would allow consumers to find their desired food product by placing constraints on the nutrition factors and ranking the obtained results based on their criteria. However, collecting nutrition information for tens of thousands of food products is time-consuming, and an automatic method is desired. This paper investigates the problem of automatic extraction of nutrition information from nutrition strings. For this purpose, it introduces a dataset of nutrition strings collected from different websites and their corresponding nutrition information. The nutrition extraction problem can be viewed as a slot filling problem, and two transformer-based methods from the literature are evaluated. The paper also introduces a specialized algorithm based on dynamic programming, and evaluates it as well as the transformer-based methods and GPT-4 and GPT-4o, with encouraging results.

Paper Type: Long

Research Area: NLP Applications

Research Area Keywords: Slot Filling, NLP, Deep Learning

Contribution Types: NLP engineering experiment, Data resources

Languages Studied: English

Submission Number: 2852