Automatic construction of arguments

Patent metadata

Field	Value
Publication number	US-10438121-B2
Application number	US-201414265408-A
Country	US
Kind code	B2
Filing date	Apr 30, 2014
Priority date	Apr 30, 2014
Publication date	Oct 8, 2019
Grant date	Oct 8, 2019

Abstract

A method comprising using at least one hardware processor for receiving a topic under consideration (TUC); providing the TUC as input to a claim function, wherein the claim function is configured to mine at least one content resource, and applying the claim function to the at least one content resource, to extract claims with respect to the TUC; and providing the TUC as input to a classification function, and applying the classification function to one or more claims of the extracted claims, to output corresponding one or more classification tags, wherein each classification tag is associated with its corresponding claim.

First claim

What is claimed is:

1. A method comprising using at least one hardware processor for:
generating a claim training dataset comprising a first content item and multiple claims, each of the multiple claims being with respect to a topic of a plurality of topics, wherein the claims are selected from the first content item by a first group of people;
generating a classification training dataset comprising a classification of each claim of the multiple claims as a pro or a con claim with respect to its corresponding topic, wherein the classification is performed by a second group of people;
machine-learning a claim function for extracting claims from a provided content resource, with respect to a provided topic under consideration (TUC), wherein the machine-learning is based on the claim training dataset;
machine-learning a classification function for classifying a provided claim with respect to a provided TUC, wherein the machine-learning is based on the classification training dataset;
receiving a new TUC as a digital text;
providing the new TUC as input to the claim function;
applying the claim function to at least one content resource, to: (a) retrieve textual passages with respect to the new TUC, and (b) extract, from the textual passages, claims with respect to the new TUC, wherein the at least one content resource is accessible through a URL (Uniform Resource Locator) and contains unstructured text in digital format; and
providing the new TUC as input to the classification function, and applying the classification function to one or more claims of the extracted claims, to output corresponding one or more classification tags, wherein:
each classification tag is associated with its corresponding claim, and
the applying of the classification function comprises classifying each of the one or more claims as a pro claim or a con claim with respect to the new TUC.

2. The method of claim 1 , further comprising using said at least one hardware processor for providing a claim of the one or more claims as input to an evidence function, wherein the evidence function is configured to mine at least one content resource, and applying the evidence function to the content resource, to extract evidence supporting and associated with the claim.

3. The method of claim 2 , wherein:
each of the extracted claims is associated with a claim score,
each of the outputted classification tags is associated with a classification score, and
each piece of evidence of the extracted evidence is associated with an evidence score,
and wherein the at least one hardware processor is further used for calculating a fused score for each of the one or more claims based on its associated claim score, classification score and evidence scores.

4. The method of claim 2 , further comprising using said at least one hardware processor for generating a list of arguments comprising a plurality of arguments, wherein each argument of the plurality of arguments comprises:
i) one different claim of the one or more claims,
ii) evidence which is associated with the one different claim, and
iii) a classification tag of the classification tags associated with the one different claim.

5. The method of claim 4 , further comprising using said at least one hardware processor for displaying the list of arguments.

6. The method of claim 5 , wherein the displaying of the list of arguments further comprises ordering the list of arguments according to the fused scores of the one or more claims corresponding to the arguments.

7. The method of claim 3 , further comprising using said at least one hardware processor for determining weights, wherein the determining of the weights comprises:
determining a claim weight for the claim scores,
determining an evidence weight for the evidence scores,
determining a classification weight for the classification scores, and
wherein the calculating of the fused score is further based on the determined weights.

8. The method of claim 4 , further comprising using said at least one hardware processor for refining at least one argument of the plurality of arguments.

9. The method of claim 4 , further comprising using said at least one hardware processor for phrasing at least one argument of the plurality of arguments.

10. A computer program product comprising a non-transitory computer-readable storage medium having program code embodied therewith, the program code executable by at least one hardware processor to:
generate a claim training dataset comprising a first content item and multiple claims, each of the multiple claims being with respect to a topic of a plurality of topics, wherein the claims are selected from the first content item by a first group of people;
generate a classification training dataset comprising a classification of each claim of the multiple claims as a pro or a con claim with respect to its corresponding topic, wherein the classification is performed by a second group of people;
machine-learn a claim function for extracting claims from a provided content resource, with respect to a provided topic under consideration (TUC), wherein the machine-learning is based on the claim training dataset;
machine-learn a classification function for classifying a provided claim with respect to a provided TUC, wherein the machine-learning is based on the classification training dataset;
receive a new TUC as a digital text;
provide the new TUC as input to the claim function;
applying the claim function to at least one content resource, to: (a) retrieve textual passages with respect to the new TUC, and (b) extract, from the textual passages, claims with respect to the new TUC, wherein the at least one content resource is accessible through a URL (Uniform Resource Locator) and contains unstructured text in digital format; and
provide the new TUC as input to the classification function, and apply the classification function to one or more claims of the extracted claims, to output corresponding one or more classification tags, wherein:
each classification tag is associated with its corresponding claim, and
the applying of the classification function comprises classifying each of the one or more claims as a pro claim or a con claim with respect to the new TUC.

11. The computer program product of claim 10 , wherein the program code is executable by the at least one hardware processor further to provide a claim of the one or more claims as input to an evidence function, wherein the evidence function is configured to mine at least one content resource, and apply the evidence function to the content resource, to extract evidence supporting and associated with the claim.

12. The computer program product of claim 11 , wherein:
each of the extracted claims is associated with a claim score,
each of the one or more claims is associated with a classification tag, and
each of the extracted piece of evidence is associated with an evidence score,
and wherein the program code is executable by the at least one hardware processor further to calculate a fused score for each of the one or more claims based on its associated claim score, classification score and evidence scores.

13. The computer program product of claim 11 , wherein the program code is executable by the at least one hardware processor further to generate a list of arguments comprising a plurality of arguments, wherein each argument of the plurality of arguments comprises:
(i) one different claim of the one or more claims,
(ii) evidence which is associated with the one different claim, and
(iii) a classification tag of the classification tags associated with the one different claim.

Assignees

• IBM

Inventors

• Aharoni Ehud
• Gutfreund Dan
• Hershcovich Daniel
• Lavee Tamar
• Levy Ran
• Rinott Ruty
• Slonim Noam
• Carmel David

Classifications

• G06N20/00

  Machine learning · CPC title

• G06N5/04 Primary

  Inference or reasoning models · CPC title