Loading…
GraphConnect 2020 has ended
Back To Schedule
Wednesday, April 22 • 1:30pm - 1:45pm
Knowledge Graphs for AI-Powered Shopping Assistants

Sign up or log in to save this to your schedule, view media, leave feedback and see who's attending!

The world around us is a network of connected concepts. Concepts can belong to different domains, but they are not isolated, rather maintain some connections with each other. Graph databases contain these connections besides the data, in contrast to traditional databases where relations are found by expensive operations during query time. In recent years, conversational systems have been trending as they provide an effortless shopping experience for customers. One of the challenges towards building a seamless dialogue with the user is understanding the products in the catalog. We address this by exploiting the perks of graph databases into the conversational systems for e-commerce through creating a knowledge graph representation for catalog data. This graph representation can be applied to many domains including grocery, movies, furniture, fashions, etc. Moreover, it can be beneficial in several facets within each domain, such as improving the training data for NLU models or answering product-related questions or even to narrow down the search results for search refinement queries.

To enable these, we first link retail concepts to catalog products, each with a specific brand, size, color, type, etc. In our application, this particular linkage between these two entities was missing initially but we utilize other hierarchical relations and design a simple but powerful semi-supervised-learning algorithm to create this linkage. To enhance our algorithm, we utilize the user logs carrying insightful information among these entities. We load the output of our algorithm into the neo4j database, a fast visual graph database supporting multiple hop query and node properties. In particular, to further benefit from Neo4j, we use its Cypher query language feature to address knowledge-based questions by transforming natural queries to Cypher queries.

Speakers
GA

Ghodrat Aalipour

Senior Data Scientist, Walmart Labs
I joined Walmart Labs in September 2018 and since then, I have been working on NLU systems for e-Commerce, please see here or here. Prior to that I was a lecturer in the School of Mathematical Sciences at RIT and a visiting faculty at the University of Colorado Denver. I have a P... Read More →


Wednesday April 22, 2020 1:30pm - 1:45pm EDT
Room 6