Information Extraction and Graph Representation for the Design of Formulated Products

Abstract

Formulated products like cosmetics, personal and household care, and pharmaceutical products are ubiquitous in everyday life. The multi-billion-dollar formulated products industry depends primarily on experiential knowledge for the design of new products. Vast knowledge of formulation ingredients and recipes exists in offline and online resources. Experts often use rudimentary searches over this data to find ingredients and construct recipes. This state of the art leads to considerable time to market and cost. We present an approach for formulated product design that enables extraction, storage, and non-trivial search of details required for product variant generation. Our contributions are threefold. First, we show how various information extraction techniques can be used to extract ingredients and recipe actions from textual sources. Second, we describe how to store this highly connected information as a graph database with an extensible domain model. And third, we demonstrate an aid to experts in putting together a new product based on non-trivial search. In an ongoing proof of concept, we use 410 formulations of various cosmetic creams to demonstrate these capabilities with promising results.