Generative Interpretable Visual Design: Using Disentanglement for Visual Conjoint Analysis
76 Pages Posted: 13 Jul 2022 Last revised: 11 Oct 2023
Date Written: October 15, 2023
Abstract
This paper develops a method to automatically discover and quantify human-interpretable visual characteristics directly from image data. Our method is generative, and creates new ``ideal point" visual designs for targeted consumer segments. The approach developed here builds on ``disentanglement" methods in deep learning using variational autoencoders, which aim to discover underlying statistically independent and interpretable characteristics constituting an object's visual representation. While the deep learning literature shows that supervision with ``ground truth" characteristics is required to obtain unique disentangled representations, these are typically unknown in real world applications and are in fact exactly the characteristics we aim to discover. Our method instead uses readily available structured product characteristics as supervisory signals to enable disentanglement, and discovers and quantifies human interpretable and statistically independent characteristics without domain knowledge on the product category. The approach is used with a dataset on watches to automatically discover interpretable visual product characteristics, obtain consumer preferences over these characteristics, and generate new ``ideal point" visual product designs targeted to specific consumer segments.
Keywords: discovery of product characteristics, deep learning, disentanglement
Suggested Citation: Suggested Citation