Abstract
The disclosed technology describes a method for programmatically predicting conceptual-level product availability, such as for brands, categories, and product lines, for stores that can lack detailed inventory feeds. A machine learning model can ingest and analyze multimodal signals, including unstructured data from user-generated content like in-store photos, crawled website, and text/images. The model processes this evidence to predict the level of confidence that a given product concept is available at a specific local store. This method can increase local product data coverage without requiring formal inventory feeds from merchants, enabling a system to surface more relevant in-store product information to consumers.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
D'Arcy, Cathal; Garnier, Nicolas; Zhao, Yiqing; Khatipova, Alisa; Minashkin, Anton; Collins, Jack; Schuler, Kathrin; Kumar, Pawan; and Ledley, Richard, "System for Modeling Local Inventory from Multimodal, Unstructured Data Signals", Technical Disclosure Commons, (July 14, 2025)
https://www.tdcommons.org/dpubs_series/8346