Entity linking via disambiguation using machine learning techniques

1. A system comprising:

a memory; and

a computer-readable medium having instructions stored thereon, which, when executed by a processor, cause the system to:

receive a text document;

for each of one or more terms in the text document, identify one or more entities to which the term potentially maps, wherein the text document includes at least one ambiguous term, an ambiguous term being a term that potentially maps to two different, but similarly named, entities;

extract one or more features from the text document;

apply an attention model to the text document based on the extracted one or more features, resulting in an attention weight being applied to each of the one or more terms in the text document;

encode the one or more terms based on the attention weights;

classify each of one or more ambiguous terms based on the encoded terms, the classification assigning a value to each different entity that each ambiguous term potentially maps to;

evaluate a minimum entropy loss function using the classification, by:

measuring entropy of an iteration of the attention model by performing a negative logarithm of a probability mass function for a value in the iteration;

determining that the measured entropy is lower than all previously measured entropies for the value in prior iterations of the attention model; and

returning an indication that the measured entropy is lower than all previously measured entropies as a result of the minimum entropy loss function;

and

back-propagate results from the minimum entropy loss function to the attention model, causing the attention model to produce different results each time it is applied until the minimum entropy loss function indicates that entropy in the classification has been minimized.

2. The system of claim 1 , wherein the encoding is performed using a recurrent neural network, and wherein the back-propagating includes back-propagating the results from the minimum entropy loss function to the recurrent neural network.

3. The system of claim 1 , wherein the classifying is performed using a feedforward neural network, and wherein the back-propagating includes back-propagating the results from the minimum entropy loss function to the feedforward neural network.

4. The system of claim 1 , wherein the attention model produces the different results each time it is applied by varying attention weights applied to the one or more features.

5. The system of claim 1 , wherein the attention model produces the different results each time it is applied by varying what the one or more features are.

6. The system of claim 1 , wherein the one or more features include term frequency-inverse document frequency (TF-IDF).

7. The system of claim 1 , wherein the one or more features include one or more features extracted from a member profile of an author of the text document.

8. A computerized method comprising:

receiving a text document;

for each of one or more terms in the text document, identifying one or more entities to which the term potentially maps, wherein the text document includes at least one ambiguous term, an ambiguous term being a term that potentially maps to two different, but similarly named, entities;

extracting one or more features from the text document;

applying an attention model to the text document based on the extracted one or more features, resulting in an attention weight being applied to each of the one or more terms in the text document;

encoding the one or more terms based on the attention weights;

classifying each of one or more ambiguous terms based on the encoded terms, the classification assigning a value to each different entity that each ambiguous term potentially maps to;

evaluating a minimum entropy loss function using the classification, by:

measuring entropy of an iteration of the attention model by performing a negative logarithm of a probability mass function for a value in the iteration;

determining that the measured entropy is lower than all previously measured entropies for the value in prior iterations of the attention model; and

returning an indication that the measured entropy is lower than all previously measured entropies as a result of the minimum entropy loss function;

and

back-propagating results from the minimum entropy loss function to the attention model, causing the attention model to produce different results each time it is applied until the minimum entropy loss function indicates that entropy in the classification has been minimized.

9. The method of claim 8 , wherein the encoding is performed using a recurrent neural network, and wherein the back-propagating includes back-propagating the results from the minimum entropy loss function to the recurrent neural network.

10. The method of claim 8 , wherein the classifying is performed using a feedforward neural network, and wherein the back-propagating includes back-propagating the results from the minimum entropy loss function to the feedforward neural network.

11. The method of claim 8 , wherein the attention model produces the different results each time it is applied by varying attention weights applied to the one or more features.

12. The method of claim 8 , wherein the attention model produces the different results each time it is applied by varying what the one or more features are.

13. The method of claim 8 , wherein the one or more features include term frequency-inverse document frequency (TF-IDF).

14. The method of claim 8 , wherein the one or more features include one or more features extracted from a member profile of an author of the text document.

15. A non-transitory machine-readable storage medium comprising instructions, which when implemented by one or more machines, cause the one or more machines to perform operations comprising:

receiving a text document;

for each of one or more terms in the text document, identifying one or more entities to which the term potentially maps, wherein the text document includes at least one ambiguous term, an ambiguous term being a term that potentially maps to two different, but similarly named, entities;

extracting one or more features from the text document;

applying an attention model to the text document based on the extracted one or more features, resulting in an attention weight being applied to each of the one or more terms in the text document;

encoding the one or more terms based on the attention weights;

classifying each of one or more ambiguous terms based on the encoded terms, the classification assigning a value to each different entity that each ambiguous term potentially maps to;

evaluating a minimum entropy loss function using the classification, by:

measuring entropy of an iteration of the attention model by performing a negative logarithm of a probability mass function for a value in the iteration;

determining that the measured entropy is lower than all previously measured entropies for the value in prior iterations of the attention model; and

returning an indication that the measured entropy is lower than all previously measured entropies as a result of the minimum entropy loss function;

and

back-propagating results from the minimum entropy loss function to the attention model, causing the attention model to produce different results each time it is applied until the minimum entropy loss function indicates that entropy in the classification has been minimized.

16. The non-transitory machine-readable storage medium of claim 15 , wherein the encoding is performed using a recurrent neural network, and wherein the back-propagating includes back-propagating the results from the minimum entropy loss function to the recurrent neural network.

17. The non-transitory machine-readable storage medium of claim 15 , wherein the classifying is performed using a feedforward neural network, and wherein the back-propagating includes back-propagating the results from the minimum entropy loss function to the feedforward neural network.

18. The non-transitory machine-readable storage medium of claim 15 , wherein the attention model produces the different results each time it is applied by varying attention weights applied to the one or more features.

19. The non-transitory machine-readable storage medium of claim 15 , wherein the attention model produces the different results each time it is applied by varying what the one or more features are.

20. The non-transitory machine-readable storage medium of claim 15 , wherein the one or more features include term frequency-inverse document frequency (TF-IDF).