Graph MayonnAIse in silico condiment synthesis Mayonnaise $12.5 - PowerPoint PPT Presentation

Graph MayonnAIse in silico condiment synthesis

Mayonnaise $12.5 billion industry by 2023 ● Regional / country-specific tastes ● Sensory profiles ● sweetness, saltiness, sourness ○ yellowiness, shininess ○ smoothness, viscosity ○ Vector representation ●

Consumer Liking Consumer surveys , e.g. 6.3 / 10 ● Country-specific regression models ● New product > regression model > predicted consumer liking ● ...

The Problem Don’t have a regression model for every country ● How to predict consumer liking in new countries? ● United States = 8.2 ○ Australia = 7.5 ○ Japan = 5.5 ○ Mexico = ? ○ Similarity-weighted sum ●

Cuisines Graph England Encode relationships between cuisines ● Online data sources ● yummly.co.uk ○ Fish & bbc.co.uk/food ○ Chips Pie & Node labels ● Mash Oil Butter 22 cuisines ○ 18,000 recipes ○ 800 ingredients ○ 1,100 flavours ○ Cod Potato Undirected edges ● Vinegar Pork Pastry

Connected Cuisines 20,000 nodes and 197,000 edges ● 329,000,000,000 cuisine-cuisine paths ●

Graph Embeddings Map nodes into a vector space ● Distance in the vector space represents structural similarity ● Machine learning algorithms ● DeepWalk ○ Node2Vec ○ TF-IDF ○

Neo4j Graph App Braintree’s flagship graph analytics platform ● Import, analyse and export data ● Run the latest machine learning algorithms with ease ● Implement algorithms in Python and run them on a Neo4j database ●

Similarity scores

Thank You For more information on the GAMMA platform please contact: gamma@braintree.com