2022
Working Paper
HBS Working Paper Series

Product2Vec: Leveraging Representation Learning to Model Consumer Product Choice in Large Assortments

By: Fanglin Chen, Xiao Liu, Davide Proserpio and Isamar Troncoso

Format:Print
| Language:English
| Pages:83

Abstract

We propose a method, Product2Vec, based on representation learning, that can automatically learn latent product attributes that drive consumer choices, to study product-level competition when the number of products is large. We demonstrate Product2Vec’s interpretability and capability for scalable causal inference. For interpretability, first, we theoretically demonstrate that there exists a direct link between product vectors and product attributes by deriving a formal proof. Second, we use product embedding to create two metrics, complementarity and exchangeability, that allow us to distinguish between products that are complements and substitutes, respectively. For causal inference, we combine product vectors with choice models and show that we can achieve better accuracy—both in terms of model fit and unbiased price coefficients—when compared to a model based solely on observable attributes, and obtain results similar to those obtained with a more complex model that includes a fixed effect for every product.