TEXTUAL ANALYSIS: IN THE AGE OF ARTIFICIAL INTELLIGENCE :
Tastelweb® allows you to process completely free expression !
Perception, qualities and defects :
Why analyze text comments?
Free comments allow an expression of the perceptions of panelists and consumers in natural language, without constraints and without filters, and whose expression contains extremely important information : rich, precise, qualitative and quantitative !
- Product reviews by experts : who has never felt a certain frustration when evaluating a product, having to limit oneself to a few intensity notes on a standardized descriptor grid ?
- Consumer reviews : open-ended questions can provide a wealth of key information about a brand or product.
Yet we have always known how to express our perceptions with words : this free language which restores all the complexity of the sensations felt when a molecule of the product, its texture or any other characteristic meets our sensory cells.
The main difficulty therefore lies in being able to process automatically and easily
this free expression without having to spend a prohibitive amount of time, to be able to group together the originalities of each individual, and not reduce too much the richness of the terms collected.
Tastelweb® allows you to process completely free expression !
1. Syntactic analysis (quantitative approach)
A sensory text remains specific through a description guided by the evolution of perception and by the nature of the words used, if we refer to a classical text.
Tastelweb® takes this specificity into account in the processing of texts from consumer tests (open questions) or descriptive tests (classic profile / QDA, Napping, Pivot© Profile, etc.).
Each sentence is syntactically analyzed in order to specify the type, gender, number, noun form of a word, etc. of each term.
Only the following will be retained : "qualifying expressions" (excluding verbal forms, adverbs, punctuation marks, etc.), with distinction between positive and negative forms.
Terms preceded, for example, by 'without', 'none', 'lack of', 'free from'… are then automatically preceded by 'non______' in order to specify their negative form.
Modalizers are also taken into account : these are words that can change the meaning of another word - example, fine bubble, absence of bubble...
The frequencies the most important qualifying expressions can be extracted and calculated per product, almost automatically.
These results make it possible to describe and compare products with each other in a completely open manner, without these terms having been previously suggested to the panelists.
2. Synthetic analysis by ChatGPT (qualitative approach)
Text comments from reviews can now be synthesized using ChatGPT version 4.0 , a version specifically customized by our IT team, for the world of organoleptic description.
As part of consumer testing, this tool integrates the precise context of the question in order to produce a relevant and adapted summary. Thanks to a personalized prompt , the panel leader only has to define the desired number of words to obtain a clear and effective summary of the comments collected.
Future, concrete developments,
thanks to Artificial Intelligence (AI)
1. Syntactic analysis (quantitative approach)
Several avenues for improvement are currently being developed :
- synonyms, antonyms: Tastelweb® already allows synonyms and antonyms to be taken into account, but manually. AI will be able to perform this grouping or dissociation work automatically, by learning products and product families.
- dictionary of words: this grouping work is now made easier by word dictionaries. The latter, specific to each product category, could be generated and personalized in an automated manner.
2. Synthetic analysis by ChatGPT (qualitative approach)
Other avenues for improvement currently being developed on this second approach :
- experts / consumers : improve the rendering of summaries according to the type of respondents.
- product family : adapt the results according to the product categories evaluated.
- comments categorization : to be able to distinguish comments according to the theme discussed by the panelists, for example comments on the presentation, taste, texture...of the products.
Copyright © All rights reserved.
Site created by ABT Informatique
5 rue du Chateau, 52300 Rouvroy-sur-Marne, France
| +33 325 94 23 00