Process for identifying completion of domain adaptation dictionary activities

Patent metadata

Field	Value
Publication number	US-10872205-B2
Application number	US-201916376338-A
Country	US
Kind code	B2
Filing date	Apr 5, 2019
Priority date	Jan 6, 2017
Publication date	Dec 22, 2020
Grant date	Dec 22, 2020

Abstract

An apparatus comprising a memory and a processor configured for semi-autonomous natural language processing domain adaptation related activities. The processor coupled to the memory and configured to identify a corpus of documents of an evaluation domain and generate a first lexicon based on the corpus of documents of the evaluation domain, and determine a threshold that indicates a sufficiency of domain adaptation of the evaluation domain based at least in part on the first lexicon. The processor is further configured to identify a corpus of documents of a client domain, generate a second lexicon based on the corpus of documents of the client domain, determine a metric associated with the corpus of documents of the client domain and the second lexicon, and determine that domain adaptation of the client domain is complete when the metric exceeds the threshold.

First claim

What is claimed is:

1. A computer-implemented method comprising:
identifying, by a processor, a corpus of documents from within a domain;
determining, by the processor, an evaluation question for use with a question answering system to determine an answer to the evaluation question based on content of the domain;
partitioning the corpus of documents into a plurality of sub-corpora;
generating a lexicon for each of the respective sub-corpora;
generating a plurality of test systems each corresponding uniquely to one of the plurality of sub-corpora;
evaluating the evaluation question using the plurality of test systems to determine a plurality of evaluation results each corresponding uniquely to one of the plurality of test systems; and
determining a threshold for sufficiency of domain adaptation based on at least one of the evaluation results.

2. The computer-implemented method of claim 1 , wherein generating the plurality of test systems each corresponding uniquely to one of the plurality of sub-corpora comprises combining one or more lexicons of each of the respective sub-corpora to generate the plurality of test systems.

3. The computer-implemented method of claim 1 , wherein evaluating the evaluation question using the test systems to determine the plurality of evaluation results each corresponding uniquely to one of the plurality of test systems comprises determining that a first evaluation result of the plurality of evaluation results corresponding to a first test system of the plurality of test systems is greater than or equal to a second evaluation result of the plurality of evaluation results that corresponds to a second test system of the plurality of test systems.

4. The computer-implemented method of claim 3 , wherein determining the threshold for sufficiency of domain adaptation based on at least one of the evaluation results comprises determining a ratio of domain terms associated with the first evaluation result and the first test system which are not associated with a third evaluation result of a third test system that is prior to the first test system to a total number of the domain terms associated with the first evaluation result and the first test system.

5. The computer-implemented method of claim 4 , wherein the threshold for sufficiency of domain adaptation is determined according to

T
=



L
n

⋂

⋃
i

n
-
1

⁢

L
i





L
n



,

wherein T is the threshold for sufficiency of domain adaptation, wherein L n is a lexicon containing the domain terms associated with the first evaluation result and the first test system, wherein ∩ denotes an intersection operation, wherein ∪ denotes a union operation, and wherein i is an index beginning at 1.

6. The computer-implemented method of claim 5 , wherein the lexicon containing the domain terms associated with the first evaluation result and the first test system contains domain terms associated with one or more tests systems prior to the first test system in the plurality of test systems.

7. The computer-implemented method of claim 1 , wherein before identifying, by the processor, the corpus of documents from within the domain, the method further comprises:
receiving a first question for processing, by the processor, according to natural language processing; and
performing, by the processor, first natural language processing to determine a first answer to the first question, wherein after determining the threshold for sufficiency of domain adaptation based on the at least one of the evaluation results, the method further comprises:
performing, by the processor, domain adaptation of a client domain to determine a second lexicon of the client domain;
receiving, by the processor, a second question for processing according to natural language processing; and
performing, by the processor, second natural language processing to determine a second answer to the second question based at least in part on the second lexicon, and wherein the second answer has a greater degree of accuracy with respect to the second question than a degree of accuracy of the first answer with respect to the first question.

8. An apparatus comprising:
a memory comprising computer-readable instructions; and
a processor coupled to the memory and configured to execute the instructions, which causes the processor to be configured to:
identify a corpus of documents from within a domain;
determine an evaluation question for use with a question answering system to determine an answer to the evaluation question based on content of the domain;
partition the corpus of documents into a plurality of sub-corpora;
generate a lexicon for each of the respective sub-corpora;
generate a plurality of test systems each corresponding uniquely to one of the plurality of sub-corpora;
evaluate the evaluation question using the plurality of test systems to determine a plurality of evaluation results each corresponding uniquely to one of the plurality of test systems; and
determine a threshold for sufficiency of domain adaptation based on at least one of the evaluation results.

9. The apparatus of claim 8 , wherein generating the plurality of test systems each corresponding uniquely to one of the plurality of sub-corpora comprises combining one or more lexicons of each of the respective sub-corpora to generate the plurality of test systems.

10. The apparatus of claim 8 , wherein evaluating the evaluation question using the test systems to determine the plurality of evaluation results each corresponding uniquely to one of the plurality of test systems comprises determining that a first evaluation result of the plurality of evaluation results corresponding to a first test system of the plurality of test systems is greater than or equal to a second evaluation result of the plurality of evaluation results that corresponds to a second test system of the plurality of test systems.

11. The apparatus of claim 10 , wherein determining the threshold for sufficiency of domain adaptation based on at least one of the evaluation results comprises determining a ratio of domain terms associated with the first evaluation result and the first test system which are not associated with a third evaluation result of a third test system that is prior to the first test system to a total number of the domain terms associated with the first evaluation result and the first test system.

12. The apparatus of claim 11 , wherein the threshold for sufficiency of domain adaptation is determined according to

T
=

Assignees

• IBM

Inventors

• Beller Charles E
• Chase Paul J
• Darden Richard
• Drzewucki Michael
• Katz Edward G
• Phipps Christopher

Classifications

• G06F16/93

  Document management systems · CPC title

• G06F40/242 Primary

  Dictionaries · CPC title

• G06F40/284 Primary

  Lexical analysis, e.g. tokenisation or collocates · CPC title

• G06F40/30

  Semantic analysis · CPC title

• G06F16/3344

  using natural language analysis · CPC title