During the last fifteen years at Repères we have developed a high level of expertise in the analysis of market research data for the modelling of consumer preferences. Our work has been based on Trade-Off analysis models and on Preference Mapping tools. From services that were only initially proposed within the framework of market studies conducted exclusively by Repères, we progressively evolved towards services involving customised modelling, by working on data from a variety of sources that our clients provided us with: sensorial evaluations, physico-chemical measurements, panel data, client data-base extractions...
In order to further develop this expertise and to promote it more widely we decided to create within Repères a department dedicated to Datamining, which intervenes either independently, based on client data, or in synergy with Repères research projects. The management of this department has been entrusted to Fabien Craignou, previously in charge of quantitative Senior studies at Repères and a specialist in research involving preference modelling.
As a complement to the Trade-Off and Preference Mapping models referred to above, the team now has extensive expertise in the use of Bayesian networks, an especially effective learning and modelling technique, that we have applied successfully for the development of typologies and scoring based on client data.
In compliance with the vision of Repères, the objective of this department is to take part in the development of new tools for the profession and the dissemination of new practises. Hence Fabien's presentation, in partnership with Lionel Jouffe from Bayesia, of our work on the identification and modelling of the appreciation of a product at the latest Sensometrics seminar in St Catharines in Canada:
Also attached is another presentation from our Datamining department, again in partnership with Bayesia, conducted during the SKIM conference in Barcelona last May. The objective was to present the advantages of Bayesian networks for defining consumer typologies, namely within the context of the ‘Uses and Attitudes’ studies. Among the advantages of the method a key point is that we analyse links of all variables 2 by 2 without establishing any principle ex ante – contrary to the usual approach of canonical typology, which requires dividing variables into two groups, for example, on one side uses and on the other attitudes, a method which is at times highly arbitrary (“I use this product every day”, is that a behaviour or an attitude, or both?).
For more details concerning these services, or more generally on Datamining services, please contact Fabien Craignou.