Grouping and linking facts from text to remove ambiguity using knowledge graphs

A system for disambiguating textual facts by linking the textual facts to a reference knowledge graph and introducing entities and relations for the textual facts, the system comprising one or more hardware processors, which, alone or in combination, are configured to provide for execution of the following steps:

• determining textual fact embeddings for the textual facts from textual data based on using a first pre-trained language model (LM),
• wherein each of the textual facts is in a triple data structure comprising a plurality of elements, wherein each of the plurality of elements is a subject, a relation, or an object;
• determining one or more knowledge graph entities and relation embedding vectors using frozen knowledge graph embeddings;
• determining one or more intermediate knowledge graph embeddings using a second pre-trained LM;
• determining knowledge graph embeddings based on using an aggregator, the one or more knowledge graph entities and relation embedding vectors, and the one or more intermediate knowledge graph embeddings;
• ranking the textual facts based on the knowledge graph embeddings and the textual fact embeddings;
• based on the ranking, determining first outputs indicating that a first subset of the plurality of elements should be linked to an existing entry within the reference knowledge graph and second outputs indicating that a second subset of the plurality of elements should be added as a new entry within the reference knowledge graph;
• and
• generating an enriched knowledge graph based on the first outputs and the second outputs.

A tangible, non-transitory computer-readable medium having instructions thereon which, upon being executed by one or more processors, alone or in combination, provide for execution of a method for disambiguating textual facts by linking the textual facts to a reference knowledge graph and introducing entities and relations for the textual facts comprising the following steps:

• determining textual fact embeddings for the textual facts from textual data based on using a first pre-trained language model (LM), wherein each of the textual facts is in a triple data structure comprising a plurality of elements, wherein each of the plurality of elements is a subject, a relation, or an object;
• determining one or more knowledge graph entities and relation embedding vectors using frozen knowledge graph embeddings;
• determining one or more intermediate knowledge graph embeddings using a second pre-trained LM;
• determining knowledge graph embeddings based on using an aggregator, the one or more knowledge graph entities and relation embedding vectors, and the one or more intermediate knowledge graph embeddings;
• ranking the textual facts based on the knowledge graph embeddings and the textual fact embeddings;
• based on the ranking, determining first outputs indicating that a first subset of the plurality of elements should be linked to an existing entry within the reference knowledge graph and second outputs indicating that a second subset of the plurality of elements should be added as a new entry within the reference knowledge graph;
• and
• generating an enriched knowledge graph based on the first outputs and the second outputs.