日本語
Conjoint analysis is widely used tool used in marketing research for new product development testing. It's usually executed as an online survey format with a few survey respondents will make 1 choice out of a set of different alternatives. The output allows researches to accurately identify what product features and combination works best before developing them.
This notebook outlines how to run a Choice-based-conjoint analysis as part of the broader Conjoint Analysis topic, with the focus being on the modelling aspect to derive preference scores. With DataRobot's SHAP values, this adds additional value of interpretability over the traditional method that is used being a linear regression coefficient scores where negative coefficients makes it hard to interpret.
From a technical prespective, in a few words as possible, conjoint analysis is a method to identify the respondents' (customers') preferences of a product feature, without explictly asking them about that product feature in a survey. This is because each respondend is forced to choose 1 item out of a set of different alternatives. Each alternative is made up of different feature combinations and permutations we are seeking to test. As we run through the survey across the large number of responses, we will identify the latent / unconcious feature preferences that the customer likes the most, that they themselves may not realise.
