Nested named entity recognition
Named Entity Recognition is the identification and classification of named entities within a document. Nested NEs occur when an NE is contained within another NE. The disclosed invention leverages the CapsNet architecture for improved nested NE identification and classification. This includes deriving the features of an input text. The derived features are used to identify and classify any named entities in the text. The system is further configured to identify named entities in the text and perform clustering to group named entities. The disclosed CapsNet considers the context of the whole text to activate higher capsule layers in order to identify the named entities and classify them. The teachings of this invention are applicable to other NER models to improve nested NE identification and classification.
1. A computer-implemented method for nested named entity recognition, comprising:
receiving, into a stack of neural capsule embedding networks comprised of M number of flat neural capsule embedding networks identified 1 through M, an embedding vector as input, wherein:
a) the embedding vector contains embedding tokens representing words in a natural language text;
b) each neural capsule embedding networks is configured to identify named entities of its assigned word span length, 1 through M;
analyzing, by each neural capsule embedding network simultaneously, the features of each word in context of the embedding vector considering tokens to the left and right of the word;
through dynamic routing of capsules, by each neural capsule embedding network simultaneously, converging to a final capsule layer mapping to each word in the input vector;
generating, by each neural capsule embedding network simultaneously, an output vector, wherein each output vector value:
a) identifies if a word span, of the neural capsule embedding network's assigned word span length, in the input is a named entity or not a named entity;
b) if the word span is a named entity, identifies what cluster the named entity belongs to.
2. The method of claim 1 further comprising:
before receiving, into a stack of neural capsule embedding networks, an embedding vector as input:
a) receiving as input a natural language text;
b) converting words in the natural language text into embeddings and inserting embedding tokens into an embedding vector.
3. The method of claim 2 , wherein converting words in the natural language text into embeddings includes populating with the value of zero any embeddings in the vector that do not correspond to a word.
4. The method of claim 1 , wherein generating, by each neural capsule embedding network, an output vector includes mathematical scaling of output vector values.
5. The method of claim 1 , further comprising:
after receiving, into a stack of neural capsule embedding networks, an embedding vector as input, deriving, by the neural capsule embedding network, features of each word in the context of the natural language text.
6. The method of claim 1 further comprising:
before receiving, into a stack of neural capsule embedding network, an embedding vector as input:
a) receiving as input a natural language text;
b) preprocessing the natural language text to identify features of the natural language text;
c) converting words in the natural language text into embeddings and inserting embedding tokens into an embedding vector.
7. The method of claim 1 , wherein clusters are a predefined set of named entity classes.
8. The method of claim 1 , wherein clusters are determined by the neural capsule embedding network.
9. The method of claim 1 further comprising:
after generating, by each neural capsule embedding network, an output vector, performing, by a neural network layer, mathematical scaling on the output vector values.
10. The method of claim 1 further comprising:
after generating, by each neural capsule embedding network, an output vector, combining all output vectors into a matrix.
11. The method of claim 10 , wherein the output vectors are combined into a matrix by:
for each non-zero integer value in the vector, inserting a 1 in each word span matrix cell, where the column is the position of each word of the word span and the row is the cluster number;
if at least one word span matrix contains a 1 at a cell location, inserting a 1 into a combined output matrix at that cell location.
12. The method of claim 10 , wherein the output vectors are combined into a matrix by inserting each vector into a row in the matrix, where the column is the position of each word of the word span and the row is the word span number.
13. A computer-implemented method for nested named entity recognition, comprising:
receiving, into a stack of named entity recognition models comprised of M number of flat named entity recognition models identified 1 through M, a natural language text as input, wherein each named entity recognition model is configured to identify named entities of its assigned word span length, 1 through M;
running all named entity recognition models simultaneously to generate an output for each named entity recognition model;
converting the output of each named entity recognition model to an output vector, wherein each output vector value:
a) identifies if a word span, of the named entity recognition model's assigned word span length, in the input is a named entity or not a named entity;
b) if the word span is a named entity, identifies what cluster the named entity belongs to.
14. The method of claim 13 further comprising:
after receiving, into a stack of named entity recognition models, an natural language text as input, converting words in the natural language text into an input that the named entity recognition model is configured to receive.
15. The method of claim 13 further comprising:
after converting the output of each named entity recognition model to an output vector, combining all output vectors into a matrix.
16. The method of claim 15 , wherein the output vectors are combined into a matrix by:
for each non-zero integer value in the vector, inserting a 1 in each word span matrix cell, where the column is the position of each word of the word span and the row is the cluster number;
if at least one word span matrix contains a 1 at a cell location, inserting a 1 into a combined output matrix at that cell location.
17. The method of claim 15 , wherein the output vectors are combined into a matrix by inserting each vector into a row in the matrix, where the column is the position of each word of the word span and the row is the word span number.
18. A system for nested named entity recognition, comprising at least one processor, the at least one processor configured to cause the system to at least perform:
receiving, into a stack of neural capsule embedding networks comprised of M number of flat neural capsule embedding networks identified 1 through M, an embedding vector as input, wherein:
a) the embedding vector contains embedding tokens representing words in a natural language text;
b) each neural capsule embedding networks is configured to identify named entities of its assigned word span length, 1 through M;
analyzing, by each neural capsule embedding network simultaneously, the features of each word in context of the embedding vector considering tokens to the left and right of the word;
through dynamic routing of capsules, by each neural capsule embedding network simultaneously, converging to a final capsule layer mapping to each word in the input vector;
generating, by each neural capsule embedding network simultaneously, an output vector, wherein each output vector value:
a) identifies if a word span, of the neural capsule embedding network's assigned word span length, in the input is a named entity or not a named entity;
b) if the word span is a named entity, identifies what cluster the named entity belongs to.
19. The system of claim 18 further comprising:
before receiving, into a stack of neural capsule embedding networks, an embedding vector as input:
a) receiving as input a natural language text;
b) converting words in the natural language text into embeddings and inserting embedding tokens into an embedding vector.
20. The system of claim 18 further comprising:
after generating, by each neural capsule embedding network, an output vector, combining all output vectors into a matrix.