N-ary relation prediction over text spans

1. A computing system, comprising:

a processor and associated memory, the processor being configured to execute one or more programs stored in the memory to:

receive an input text;

process the input text using a neural network system, the neural network system including at least a first neural network and a second neural network, the neural network system having been trained using a training data set;

wherein to process the input text, the processor is configured to, for each of a plurality of text spans within the input text:

identify a vector of a plurality of semantic entities;

identify a vector of entity mentions for each of the semantic entities;

define an n-ary relation between entity mentions of the semantic entities, the n-ary relation having an associated plurality of subrelations;

determine one or more mention-level representation vectors in each of the plurality of text spans that satisfy at least one of the n-ary relation or subrelations for each of the semantic entities using the first neural network;

wherein to process the input text, the processor is further configured to, for the input text, including all of the text spans:

aggregate the mention-level representation vectors over all of the text spans to thereby produce entity-level representation vectors for the input text;

input, to the second neural network, the entity-level representation vectors; and

responsive to the input of the entity-level representation vectors, output from the second neural network a prediction of a presence of the n-ary relation for the semantic entities in the input text.

2. The system of claim 1 , wherein the text spans are discontiguous.

3. The system of claim 1 , wherein the text spans are longer than a sentence.

4. The system of claim 1 , wherein the plurality of text spans includes text spans at a plurality of scales including one or more of sentence and paragraph.

5. The system of claim 1 , wherein the input text includes an entire document having a plurality of paragraphs.

6. The system of claim 1 , further comprising the processor being configured to concatenate the entity-level representation vectors for each subrelation before inputting to the second neural network to output the prediction.

7. The system of claim 1 , wherein the prediction is an indicator of a probability of each of the n-ary relations being mapped to the semantic entities within the input text.

8. The system of claim 1 , wherein aggregating includes an aggregation operator performing an operation that includes at least one of max pooling, max pooling with smoothing, and ensemble prediction.

9. The system of claim 1 , wherein the first neural network includes a bidirectional long short-term memory recurrent neural network with a single-layer neural network, and the second neural network includes a two-layer feedforward neural network.

10. The system of claim 1 , wherein training the neural network system includes at least one of supervised training and distant supervision training.

11. The system of claim 1 , further comprising the processor being configured to test the neural network system, wherein testing includes outputting a test prediction for each text span and combining the test predictions over the plurality of text spans using an ensemble operator to generate a document-level prediction for the input text.

12. The system of claim 1 , wherein identifying the vector of entity mentions includes, at a pre-processing stage, the processor being configured to generate a mapping of a first entity mention to a second entity mention from tokenizing a document dataset and applying of at least one of a set of rules, the set of rules including at least a first predetermined rule that the first entity mention and the second entity mention are in a same token.

13. A method for use with a computing device including a processor, comprising:

receiving input text at the computing device;

processing the input text using a neural network system executed via the processor of the computing device, the neural network system including at least a first neural network and a second neural network, the neural network system having been trained using a training data set;

wherein the processing includes, for each of a plurality of text spans within the input text:

identifying a vector of a plurality of semantic entities;

identifying a vector of entity mentions for each of the semantic entities;

defining an n-ary relation between entity mentions of the semantic entities, the n-ary relation having an associated plurality of subrelations;

determining one or more mention-level representation vectors in each of the plurality of text spans that satisfy at least one of the n-ary relation or subrelations for each of the semantic entities using the first neural network;

wherein the processing further includes, for the input text, including all of the text spans:

aggregating the mention-level representation vectors over all of the text spans to thereby produce entity-level representation vectors for the input text;

inputting, to the second neural network, the entity-level representation vectors; and

responsive to the input of the entity-level representation vectors, outputting from the second neural network a prediction of a presence of the n-ary relation for the semantic entities in the input text.

14. The method of claim 13 , wherein the text spans are discontiguous.

15. The method of claim 13 , wherein the text spans are longer than a sentence.

16. The method of claim 13 , wherein the plurality of text spans includes text spans at a plurality of scales including one or more of sentence and paragraph.

17. The method of claim 13 , wherein the input text includes an entire document having a plurality of paragraphs.

18. The method of claim 13 , further comprising, at the processor, concatenating the entity-level representation vectors for each subrelation before inputting to the second neural network to output the prediction.

19. The method of claim 13 , wherein the prediction is an indicator of a probability of each of the n-ary relations being mapped to the semantic entities within the input text.

20. A method for use with a computing device including a processor, comprising:

providing a neural network system executed by the processor of the computing device, the neural network system including at least a first neural network that is a bidirectional long short-term memory recurrent neural network with a single-layer neural network, and a second neural network that is a two-layer feedforward neural network, the neural network system having been trained using a training data set;

receiving input text;

for each of a plurality of text spans within the input text;

identifying a vector of a plurality of semantic entities;

identifying a vector of entity mentions for each of the semantic entities;

defining an n-ary relation between entity mentions of the semantic entities, the n-ary relation having an associated plurality of subrelations;

determining one or more mention-level representation vectors in the text span that satisfy at least one of the n-ary relation or subrelations for each of the semantic entities in each of the plurality of text spans using the first neural network;

for the input text, including all of the text spans;

aggregating the mention-level representation vectors over all of the text spans to thereby produce entity-level representation vectors for the input text;

inputting, to the second neural network, the entity-level representation vectors; and

responsive to the input of the entity-level representation vectors, outputting from the second neural network a prediction of a presence of the n-ary relation for the semantic entities in the input text,

wherein the text spans are discontiguous.