Hierarchical relationship extraction

An apparatus comprising:

• one or more computer processors;
• a computer-processor readable storage medium;
• an interface via which a portion of text can be received;
• and
• computer processor executable instructions stored in the storage medium that when executed by the one or more computer processors cause the apparatus to:
• receive a portion of text comprising data representative of a relationship for the one or more entities of interest, wherein the portion of text comprises multiple separable entities including one or more relationship entities and the one or more entities of interest;
• for each of the multiple separable entities, generate a set of embeddings by (a) retrieving, from an embedding vocabulary dataset, one or more embeddings of entities associated with the separable entity and (b) forming a set of embeddings associated with the separable entity based on the retrieved one or more embeddings, wherein each set of embeddings comprises an embedding of the separable entity and at least one embedding of an entity associated with the separable entity;
• send at least one embedding from each of the sets of embeddings for input to a machine learning model or classifier;
• and
• store the generated sets of embeddings in the embedding vocabulary dataset, wherein the embedding vocabulary dataset comprises data representative of one or more entities mapped to one or more corresponding embeddings,
• wherein,
• retrieving from the embedding vocabulary dataset one or more embeddings of entities associated with a separable entity further comprises (a) determining whether an embedding corresponding to each of the separable entity and the one or more entities associated with the separable entity exists in the embedding vocabulary dataset, (b) retrieving those embeddings associated with the separable entity that exist in the embedding vocabulary dataset, (c) generating out-of-vocabulary embeddings for those embeddings associated with the separable entity that are not found in the embedding vocabulary dataset, and (d) generating a set of embeddings for the separable entity based on at least any retrieved embedding or any generated out-of-vocabulary embedding.

A non-transitory computer processor readable storage medium comprising computer processor executable instructions stored in the storage medium that when executed by one or more computer processors cause an apparatus to:

• receive a portion of text comprising data representative of a relationship for the one or more entities of interest, wherein the portion of text comprises multiple separable entities including one or more relationship entities and the one or more entities of interest;
• for each of the multiple separable entities, generate a set of embeddings by (a) retrieving, from an embedding vocabulary dataset, one or more embeddings of entities associated with the separable entity and (b) forming a set of embeddings associated with the separable entity based on the retrieved one or more embeddings, wherein each set of embeddings comprises an embedding of the separable entity and at least one embedding of an entity associated with the separable entity;
• send at least one embedding from each of the sets of embeddings for input to a machine learning model or classifier;
• and
• store the generated sets of embeddings in the embedding vocabulary dataset, wherein the embedding vocabulary dataset comprises data representative of one or more entities mapped to one or more corresponding embeddings,
• wherein,
• retrieving from the embedding vocabulary dataset one or more embeddings of entities associated with a separable entity further comprises (a) determining whether an embedding corresponding to each of the separable entity and the one or more entities associated with the separable entity exists in the embedding vocabulary dataset, (b) retrieving those embeddings associated with the separable entity that exist in the embedding vocabulary dataset, (c) generating out-of-vocabulary embeddings for those embeddings associated with the separable entity that are not found in the embedding vocabulary dataset, and (d) generating a set of embeddings for the separable entity based on at least any retrieved embedding or any generated out-of-vocabulary embedding.