Beyond Structure-Property Relationships: A Correlating Customer Reviews to Consumer Product Formulations

Multi-billion dollar industries such as pharmaceuticals, food, and consumer products rely on formulations with complex performance criteria. Relating the features of the formulation to consumer perception of the product is a significant challenge. This challenge arises, because both intrinsic and extrinsic physical and chemical descriptors, as well as personal and social factors affect public perception. User ratings and individual customer feedback have been used in many social platforms to improve and personalize user experience. We hypothesize that this feedback can also be used to guide manufacturers as well. To achieve this, a workflow relating ingredients and customer satisfaction must be established. We have developed a process and tools for ingesting, curating, modeling, analyzing and visualizing consumer review data from very large online retail distributors. Our hypothesis is that the intrinsic aspects of the formulated product, i.e. composition, and product performance - both of which are available from the retailer can be used to elucidate which if any correlations between the two exist.

Customer review data - including the list of ingredients for a wide range of consumer products are accessible online. This large array of feedback and product performance defines a hyperdimensional space of public perception of the product. Similarly, the product list also provides enough information to generate a list of CAS numbers. With some restrictions, CAS numbers can be used to generate molecular representations. From the CAS, a SMILES can be matched and used to make fundamental calculations of the molecular features associated with the product. These arrays of molecular descriptors are the molecular structure space. We present a path for relating the molecular and performance spaces of a wide range of retail consumer products sold online in order to enable advanced regression tools.