Custom tags based on word embedding vector spaces

1. A non-transitory machine-readable medium storing a program executable by at least one processing unit of a device, the program comprising sets of instructions for:

receiving a plurality of sets of words, each set of words in the plurality of sets of words comprising a word annotated as being an entity having a same custom entity type, wherein the custom entity type is a custom-defined tag within a word embedding space, wherein the word embedding space is an n-dimensional vector space;

determining a plurality of word embeddings in the word embedding space for the plurality of annotated words, wherein at least one of the plurality of word embeddings in the word embedding space is a first vector representation of at least one of the annotated words;

receiving a set of words;

determining a word embedding for a subset of the set of words, wherein the word embedding is second vector representation of the subset of the set of words;

determining whether the word embedding falls within the defined region in the word embedding space; and

upon determining that the word embedding falls within the defined region in the word embedding space, determining that the subset of the set of words represents an entity having the custom entity type, wherein the entity is capable of having a plurality of custom-defined tags comprising at least the custom-defined tag.

2. The non-transitory machine-readable medium of claim 1 , wherein the plurality of word embeddings is a first plurality of word embeddings, wherein the custom entity type is a first custom entity type, wherein the region in the word embedding space is a first region in the word embedding space, wherein the program further comprises sets of instructions for:

receiving a second plurality of word embeddings in the word embedding space, each word embedding in the second plurality of word embeddings associated with a second custom entity type; and

defining a second region in the word embedding space based on the received second plurality of word embeddings.

3. The non-transitory machine-readable medium of claim 2 , wherein the entity is a first entity, wherein the program further comprises sets of instructions for:

determining whether the word embedding falls within the second defined region in the word embedding space;

upon determining that the word embedding falls within the second defined region in the word embedding space, determining that the subset of the set of words represents a second entity having the second custom entity type.

4. The non-transitory machine-readable medium of claim 1 , wherein defining the region in the word embedding space comprises generating a convex hull in the word embedding space based on the received plurality of word embeddings.

5. The non-transitory machine-readable medium of claim 4 , wherein determining whether the word embedding falls within the defined region in the word embedding space comprises determining whether the word embedding falls within a defined threshold distance of the convex hull.

6. The non-transitory machine-readable medium of claim 1 , wherein the set of words comprises raw unstructured text from a document in a medical record of a patient.

7. The non-transitory machine-readable medium of claim 1 , wherein the set of words comprises a set of words included in a textual description of a concept for an entry in a knowledge base.

8. A method comprising:

receiving a plurality of sets of words, each set of words in the plurality of sets of words comprising a word annotated as being an entity having a same custom entity type, wherein the custom entity type is a custom-defined tag within a word embedding space, wherein the word embedding space is an n-dimensional vector space;

determining a plurality of word embeddings in the word embedding space for the plurality of annotated words, wherein at least one of the plurality of word embeddings in the word embedding space is a first vector representation of at least one of the annotated words;

defining a region in the word embedding space based on the received plurality of word embeddings;

receiving a set of words;

determining a word embedding for a subset of the set of words, wherein the word embedding is second vector representation of the subset of the set of words;

determining whether the word embedding falls within the defined region in the word embedding space; and

upon determining that the word embedding falls within the defined region in the word embedding space, determining that the subset of the set of words represents an entity having the custom entity type, wherein the entity is capable of having a plurality of custom-defined tags comprising at least the custom-defined tag.

9. The method of claim 8 , wherein the plurality of word embeddings is a first plurality of word embeddings, wherein the custom entity type is a first custom entity type, wherein the region in the word embedding space is a first region in the word embedding space, wherein the method further comprises:

receiving a second plurality of word embeddings in the word embedding space, each word embedding in the second plurality of word embeddings associated with a second custom entity type; and

defining a second region in the word embedding space based on the received second plurality of word embeddings.

10. The method of claim 9 , wherein the entity is a first entity, wherein the method further comprises:

determining whether the word embedding falls within the second defined region in the word embedding space;

upon determining that the word embedding falls within the second defined region in the word embedding space, determining that the subset of the set of words represents a second entity having the second custom entity type.

11. The method of claim 8 , wherein defining the region in the word embedding space comprises generating a convex hull in the word embedding space based on the received plurality of word embeddings.

12. The method of claim 11 , wherein determining whether the word embedding falls within the defined region in the word embedding space comprises determining whether the word embedding falls within a defined threshold distance of the convex hull.

13. The method of claim 8 , wherein the set of words comprises raw unstructured text from a document in a medical record of a patient.

14. The method of claim 8 , wherein the set of words comprises a set of words included in a textual description of a concept for an entry in a knowledge base.

15. A system comprising:

a set of processing units; and

a non-transitory machine-readable medium storing instructions that when executed by at least one processing unit in the set of processing units cause the at least one processing unit to:

receive a plurality of sets of words, each set of words in the plurality of sets of words comprising a word annotated as being an entity having a same custom entity type, wherein the custom entity type is a custom-defined tag within a word embedding space, wherein the word embedding space is an n-dimensional vector space;

determine a plurality of word embeddings in the word embedding space for the plurality of annotated words, wherein at least one of the plurality of word embeddings in the word embedding space is a first vector representation of at least one of the annotated words;

define a region in the word embedding space based on the received plurality of word embeddings;

receive a set of words;

determine a word embedding for a subset of the set of words, wherein the word embedding is second vector representation of the subset of the set of words;

determine whether the word embedding falls within the defined region in the word embedding space; and

upon determining that the word embedding falls within the defined region in the word embedding space, determine that the subset of the set of words represents an entity having the custom entity type, wherein the entity is capable of having a plurality of custom-defined tags comprising at least the custom-defined tag.

16. The system of claim 15 , wherein the plurality of word embeddings is a first plurality of word embeddings, wherein the custom entity type is a first custom entity type, wherein the region in the word embedding space is a first region in the word embedding space, wherein the instructions further cause the at least one processing unit to:

receive a second plurality of word embeddings in the word embedding space, each word embedding in the second plurality of word embeddings associated with a second custom entity type; and

define a second region in the word embedding space based on the received second plurality of word embeddings.

17. The system of claim 16 , wherein the entity is a first entity, wherein the instructions further cause the at least one processing unit to:

determine whether the word embedding falls within the second defined region in the word embedding space;

upon determining that the word embedding falls within the second defined region in the word embedding space, determine that the subset of the set of words represents a second entity having the second custom entity type.

18. The system of claim 15 , wherein defining the region in the word embedding space comprises generating a convex hull in the word embedding space based on the received plurality of word embeddings.

19. The system of claim 18 , wherein determining whether the word embedding falls within the defined region in the word embedding space comprises determining whether the word embedding falls within a defined threshold distance of the convex hull.

20. The system of claim 15 , wherein the set of words comprises raw unstructured text from a document in a medical record of a patient.