Quantifying and Visualizing the Demand and Supply Gap from E-commerce Search Data using Topic Models

Abstract

The demand generation and assortment planning are two critical components of running a retail business. Traditionally, retail companies use the historical sales data for modeling and optimization of assortment selection, and they use a marketing strategy for demand generation. However, today, most retail businesses have e-commerce sites with rapidly growing online sales. An e-commerce site typically has to maintain a large amount of digitized product data, and it also keeps a vast amount of historical customer interaction data that includes search, browse, click, purchase and many other different interactions. In this paper, we show how this digitized product data and the historical search logs can be used in understanding and quantifying the gap between the supply and demand side of a retail market. This gap helps in making an effective strategy for both demand generation and assortment selection. We construct topic models of the historical search queries and the digitized product data from the catalog. We use the former to model the customer demand and the later to model the supply side of the retail business. We then create a tool to visualize the topic models to understand the differences between the supply and demand side. We also quantify the supply and demand gap by defining a metric based on Kullback-Leibler (KL) divergence of topic distributions of queries and the products. The quantification helps us identifying the topics related to excess or less demand and thereby in designing effective strategies for demand generation and assortment selection. Application of this work by e-Commerce retailers can result in the development of product innovations that can be utilized to achieve economic equilibrium. We can identify the excess demand and can provide insight to the teams responsible for improving assortment and catalog quality. Similarly, we can also identify excess supply and can provide that intelligence to the teams responsible for demand generation. Tools of this nature can be developed to systematically drive efficiency in achieving better economic gains for the entire e-commerce engine. We conduct several experiments collecting data from Walmart.com to validate the effectiveness of our approach.