Applying Text Mining Methods for Sensory Evaluation Research
Organizers: Sébastien Lê, Jacob Lahne
Participants: Jacob Lahne, Alexiane Luc, Benjamin Mahieu, Leticia Vidal
Facilitator: Sébastien Lê, Anne Hasted
Jacob Lahne, Web scraping for sensory research: a case study with cocktails.
Websites that review or describe food products are potentially rich sources of sensory data. However, many sensory scientists find acquiring and parsing web data a challenging barrier. This talk introduces web scraping for sensory data using a case study of a website with ~4500 well-structured cocktail recipes. A workflow for acquiring and parsing the data is presented (in R) with a focus on exposing the basics of interacting with HTML trees (using R). The results are analysed using featural (network) approaches to make the sensory aspects of the data visible.
Benjamin Mahieu, Sensory characterization of home-perfumes using Free-Comment as response to open-ended questions.
This talk explores text data collected using a Free-Comment protocol in which 88 consumers evaluated 4 home-perfumes by answering the question “Describe the olfactory characteristics of this perfume”. The talk presents data pre-processing (cleaning, lemmatization, etc.) and proposes an original approach based on a classification using chi-square distance to group descriptors that are “close enough” to be associated. Outputs of the statistical analyses of the pre-processed data are presented to show the relevance of the Free-Comment methodology in discriminating and characterizing a set of products.
Alexiane Luc, NLP strategies to analyse consumer data using valency: an application to Free JAR data. This talk presents the analysis of consumer free comment with a constraint of expression, which appears in the systematic use of a JAR structure to describe the tested products. This type of data is also known as Free JAR data. A new algorithm which considers all the specificities of Free JAR consumer data to understand the opinion generated by the tested products among the subjects, will be presented.
Leticia Vidal, Automatic text analysis of Twitter data. Social media is a valuable source to study consumer behaviour. Some years ago, ~50,000 tweets containing the words “breakfast”, “lunch”, “snack” or “dinner” were retrieved to investigate “what people say when tweeting about eating situations”. Content analysis was used, but due to the time-consuming nature of manual coding it was only applied to a subset of 16,000 tweets. In this talk, the use of automatic text analysis tools to gain valuable insights based on the whole dataset will be explored.
Joint SSP and sensometrics WORKSHOP:
Artificial Intelligence in Sensory Practice:
Separating Promise from Hype
Lead by Rafal Drabek
Participants: Amanda Grzeda, John Ennis, Leah Hamilton
The rapid evolution of computational technology has sparked excitement in the ability of artificial intelligence (AI) to provide never-before-considered solutions and insights. The nebulous nature of the term “AI,” along with the deep technical knowledge AI seemingly requires, has resulted in high expectations – what can’t AI do? And, while most agree the true power of AI will be unleashed when humans and machines work seamlessly together - each leveraging the strengths of the other - how will that goal be realized in sensory practice? This workshop separates the promise from the hype.
In Part 1, Amanda Grzeda shares within-business experiences. First, how to effectively translate AI excitement into strong business questions. Next, how to source the right domain knowledge, which likely lies across multiple individuals who represent different departments and functions. Then, how to cast the net wide enough with regard to metadata - considering what’s relevant today, and what might be relevant in the future. Finally, with a model in-hand, how to best play the role of the human in the human-machine interface. Data modeling is not new to us, but given the focused attention, our ability to communicate it effectively must evolve.
In Part 2, John Ennis discusses how, whenever a new general-purpose technology - such as steam, electricity, digital computing, or AI - appears, industrialists must decide how to transform. One option, “fast caterpillar,” is to continue in one’s present activities, leveraging the new technology to increase output. The second option, “beautiful butterfly,” is to use the new technology to produce results previously unattainable. In this part, Dr. Ennis reviews how AI makes both options available to sensory scientists and provides recommendations on how to balance these options to provide business benefits while effectively preparing for the future.
In Part 3, Leah Hamilton discusses opportunities and challenges in incorporating existing AI technologies into novel solutions for sensory science. Language-standardization training or analysis is often the bottleneck in sensory research, making the rapid analysis of thousands of words of natural language using AI an exciting prospect. There are great opportunities and challenges in using existing tools for novel sensory applications, as demonstrated through a case study that used descriptors from 6,598 reviews of international whiskies to create a flavor wheel. This talk highlights the power of using domain expertise to adapt existing AI, rather than reinventing the wheel.
The workshop concludes with panel discussion and open Q&A.