This automatic system of keyword recommendations for taxonomy enrichment effort was carried out to improve our customer journey leading to better customer conversion on the products resulting in incremental business impact and $64M value realization. The solution developed uses a novel Machine learning algorithms of weighted bipartite graph structure, the Louvain community detection algorithm, and its application to enrich the site taxonomy. This solution was merited for its novelty by Walmart's patent committee and is now filed for US patent.
This solution is called Automated Asset Item Recommender (AIR) which programmatically maps the right items to the assets using from millions of combinations. It uses the computer vision algorithm to map the content and come up with personalized recommendation and improve customer engagement with both inspiration and promotional content of Walmart.
This invention brings together various traditional and new transfer learning algorithms and tweaks Transformer architecture to achieve results on downstream language tasks while at the same time generalizing across business verticals and languages.
Harvard Publication
Harvard Publication
