Data from online product reviews offers a novel avenue for the sensory characterisation of food. But, little research has investigated the quality of sensory information in the online reviews. The aim of this research was to investigate consumer online reviews as sources of sensory attributes of food products (starting with a minimum of 1000 online reviews per product), and to assess how the resulting sensory product profiles compare to profiles obtained in a central location test with 105 consumers using rate-all-that-apply (RATA) questions. A case study was conducted with five unflavoured coffee samples. A semi-automated approach, combining natural language processing and sensory science expertise was used to clean online review data, develop a sensory lexicon, and analyse the frequency of attributes used by consumers. It was possible to develop online review-based sensory product profiles and discriminate the five samples on this basis. Consumers used a small set of broad, mostly intensity related sensory terms (e.g. ‘Bold/rich’, ‘Strong/intense’, ‘Smooth’, ‘Weak/bland’) more frequently than descriptive terms. Canonical analysis showed high agreement between new method and RATA for product discrimination and between two group of descriptors. The first group (including ‘Coffee Flavor,’ ‘Rich in Flavor’, and ‘Smell of Coffee’) is associated with the intensity of flavour of coffee, while the second group describes characteristic flavour of coffee (including ‘Bold’, ‘Dark’ ‘Body’). Furthermore, care should be taken when implementing these findings in food categories with lower levels of consumer engagement, where consumer comments relating to sensory properties in online reviews may be less frequent.
Read full abstract