KG Embeddings - Elastic Search

 Video -> https://youtu.be/J-v_rdyRSK8

Knowledge graphs (KGs) have become ubiquitous, created and used by many organisations (from enterprises, research institutions to crowd-sourced open communities) as backbone and driver of many of their intelligent products and applications. Many KGs such as DBpedia, Microsoft Graph, Google Knowledge Graph, Facebook Social Graph (to name few) have been created to enable and support real world use cases and downstream tasks such as recommendations systems included content-based ones. Leveraging Knowledge graph content, structure and semantics to build an entity recommender system is challenging because of the graph complexity, its high-dimensional and symbolic nature. However, recent Knowledge graph embeddings approaches have been developed to address the aforementioned challenges. The key idea here is to project KGs entities and relations into low-dimensional vector spaces simpler to deal with and compute from while preserving (as much as possible) the knowledge graph semantics, structure and entity relative similarities. Equipped with such embeddings, knowledge graph can be projected and deployed into ElasticSearch to build large-scale content-based recommender system. In this talk, I will give an introduction to Knowledge graph through real world examples, expose the challenges they raise, present different Knowledge graph embeddings techniques and demonstrate how to build, deploy and use them in ElasticSearch for entity content-based entity recommendation and similarity computation.

 

Conceptual Layer - Entity types (classes and sub classes)

Entity Layer - IDs (conected with its entity type and other entities)

Attributes Layer

JSON-LD

Data Integration

 

Elastic Search and Embeddings indexing

https://www.elastic.co/pt/blog/text-similarity-search-with-vectors-in-elasticsearch