Infusing knowledge into natural language processing tasks using graph structures
One embodiment of the invention provides a method for natural language processing (NLP). The method comprises extracting knowledge outside of text content of a NLP instance by extracting a set of subgraphs from a knowledge graph associated with the text content. The set of subgraphs comprises the knowledge. The method further comprises encoding the knowledge with the text content into a fixed size graph representation by filtering and encoding the set of subgraphs. The method further comprises applying a text embedding algorithm to the text content to generate a fixed size text representation, and classifying the text content based on the fixed size graph representation and the fixed size text representation.
1. A method for natural language processing (NLP), comprising:
extracting knowledge outside of text content of a NLP instance by extracting a set of subgraphs from a knowledge graph associated with the text content, wherein the set of subgraphs comprises the knowledge;
encoding the knowledge with the text content into a fixed size graph representation by filtering and encoding the set of subgraphs, wherein the fixed size graph representation includes at least one filtered subgraph comprising only nodes that satisfy a pre-determined threshold;
applying a text embedding algorithm to the text content to generate a fixed size text representation; and
classifying the text content based on the fixed size graph representation and the fixed size text representation.
2. The method of claim 1 , wherein the knowledge graph is one of a knowledge base, a semantic network, or a social graph.
3. The method of claim 1 , wherein the text content comprises one or more text samples.
4. The method of claim 3 , wherein the one or more text samples include a premise and a hypothesis.
5. The method of claim 3 , wherein the knowledge graph comprises one of a directed graph representation or an undirected graph representation of the one or more text samples.
6. The method of claim 1 , wherein the set of subgraphs is encoded via a Relational Graph Convolutional Network (R-GCN).
7. The method of claim 1 , wherein the set of subgraphs is filtered based on a personalized page rank (PPR) algorithm.
8. The method of claim 1 , wherein the text content is classified via a Feed Forward Network (FFN).
9. The method of claim 1 , wherein the text embedding algorithm comprises one of Bidirectional Encoder Representations from Transformers (BERT) or Global Vectors (GloVe).
10. A system for natural language processing (NLP), comprising:
at least one processor; and
a non-transitory processor-readable memory device storing instructions that when executed by the at least one processor causes the at least one processor to perform operations including:
extracting knowledge outside of text content of a NLP instance by extracting a set of subgraphs from a knowledge graph associated with the text content, wherein the set of subgraphs comprises the knowledge;
encoding the knowledge with the text content into a fixed size graph representation by filtering and encoding the set of subgraphs, wherein the fixed size graph representation includes at least one filtered subgraph comprising only nodes that satisfy a pre-determined threshold;
applying a text embedding algorithm to the text content to generate a fixed size text representation; and
classifying the text content based on the fixed size graph representation and the fixed size text representation.
11. The system of claim 10 , wherein the knowledge graph is one of a knowledge base, a semantic network, or a social graph.
12. The system of claim 10 , wherein the text content comprises one or more text samples.
13. The system of claim 12 , wherein the one or more text samples include a premise and a hypothesis.
14. The system of claim 12 , wherein the knowledge graph comprises one of a directed graph representation or an undirected graph representation of the one or more text samples.
15. The system of claim 10 , wherein the set of subgraphs is encoded via a Relational Graph Convolutional Network (R-GCN).
16. The system of claim 10 , wherein the set of subgraphs is filtered based on a personalized pagerank (PPR) algorithm.
17. The system of claim 10 , wherein the text content is classified via a Feed Forward Network (FFN).
18. The system of claim 10 , wherein the text embedding algorithm comprises one of Bidirectional Encoder Representations from Transformers (BERT) or Global Vectors (GloVe).
19. A computer program product for natural language processing (NLP), the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:
extracting knowledge outside of text content of a NLP instance by extracting a set of subgraphs from a knowledge graph associated with the text content, wherein the set of subgraphs comprises the knowledge;
encoding the knowledge with the text content into a fixed size graph representation by filtering and encoding the set of subgraphs, wherein the fixed size graph representation includes at least one filtered subgraph comprising only nodes that satisfy a pre-determined threshold;
applying a text embedding algorithm to the text content to generate a fixed size text representation; and
classifying the text content based on the fixed size graph representation and the fixed size text representation.
20. The computer program product of claim 19 , wherein the knowledge graph is one of a knowledge base, a semantic network, or a social graph.