Performance of time intensive question processing in a cognitive system

Patent metadata

Field	Value
Publication number	US-10956824-B2
Application number	US-201615372476-A
Country	US
Kind code	B2
Filing date	Dec 8, 2016
Priority date	Dec 8, 2016
Publication date	Mar 23, 2021
Grant date	Mar 23, 2021

Abstract

Mechanisms are provided for performing watched question processing. The mechanisms detect a change in content of electronic documents in a corpus of electronic documents, and retrieve a watched question from a watched question database. The mechanisms determine whether full processing of the watched question through the question answering pipeline is to be bypassed when performing watched question processing in response to the detected change in content of the electronic documents in the corpus. The mechanisms bypass one or more stages of processing of the question answering pipeline in response to determining that full processing of the watched question through the question answering pipeline is to be bypassed. The mechanisms return a result of the watched question processing indicating that no changes to the previously generated answers are caused by the change in the content of the electronic documents in the corpus.

First claim

What is claimed is:

1. A method, in a data processing system comprising a processor and a memory, the memory comprising instructions which are executed by the processor to cause the processor to implement the method, comprising:
detecting, by the data processing system, a change in content of electronic documents in a corpus of electronic documents;
retrieving, by the data processing system, a watched question from a watched question database, wherein the watched question is a question previously processed by a question answering pipeline of the data processing system and for which monitoring is performed to identify changes to previously generated answers for the watched question;
determining, by the data processing system, whether full processing of the watched question through the question answering pipeline is to be bypassed when performing watched question processing in response to the detected change in content of the electronic documents in the corpus;
in response to determining that full processing of the watched question through the question answering pipeline is to be bypassed:
bypassing, by the data processing system, one or more stages of processing of the question answering pipeline, and
returning, by the data processing system, a result of the watched question processing indicating that no changes to the previously generated answers are caused by the change in the content of the electronic documents in the corpus; and

in response to determining that full processing of the watched question through the question answering pipeline is not to be bypassed:
performing full processing of the watched question through all stages of the question answering pipeline to generate one or more new answers to the watched question based on the updated corpus;

wherein determining whether full processing of the watched question through the question answering pipeline is to be bypassed comprises:
obtaining first results comprising a first predetermined number of highest ranked documents returned by a primary search of the corpus prior to the change in content of electronic documents in the corpus;
obtaining second results comprising a second predetermined number of highest ranked documents returned by a primary search of the updated corpus after the change in content of electronic documents in the corpus;
comparing the first results to the second results; and
determining whether full processing of the watched question through the question answering pipeline is to be bypassed based on results of the comparison.

2. The method of claim 1 , further comprising determining that full processing of the watched question through the question answer pipeline is to be bypassed in response to results of the comparison indicating that the first results are equivalent to the second results.

3. The method of claim 1 , wherein the first results are obtained from a results cache, where the first results are results of a prior full processing of the watched question through the question answer pipeline.

4. The method of claim 1 , wherein obtaining second results comprising a second predetermined number of highest ranked documents returned by a primary search of the updated corpus after the change in content of electronic documents in the corpus comprises obtaining the second results based on processing of only the electronic documents in the corpus whose content was changed.

5. The method of claim 4 , wherein obtaining second results further comprises distinguishing changed portions of the corpus from previously existing portions of the corpus based on date filtering, and wherein determining whether full processing of the watched question through the question answering pipeline is to be bypassed comprises determining whether the changed portions generate a change in previously generated answers to the watched question.

6. The method of claim 4 , wherein obtaining second results further comprises distinguishing changed portions of the corpus from previously existing portions of the corpus based on a metadata facet value associated with each of the documents in the corpus indicating whether or not a corresponding document in the corpus is a changed or new document, or a previously existing document, and wherein determining whether full processing of the watched question through the question answering pipeline is to be bypassed comprises determining whether the changed portions generate a change in previously generated answers to the watched question.

7. The method of claim 1 , further comprising:
performing, by the data processing system, an operation that groups watched questions having at least one similar characteristic into a watched question pool;
selecting, by a data processing system, from the watched question pool, a watched question as a representative watched question of the watched question pool; and
replicating a result of processing the representative watched question by the data processing system to other watched questions in the watched question pool.

8. The method of claim 1 , further comprising:
determining, by the data processing system, whether a component of the question answering pipeline has been updated; and
in response to determining that a component of the question answering pipeline has been updated, marking, by the data processing system, all watched questions in the watched question database as requiring full processing through the question answering pipeline.

9. A computer program product comprising a computer readable storage medium having a computer readable program stored therein, wherein the computer readable program, when executed on a data processing system, causes the data processing system to:
detect a change in content of electronic documents in a corpus of electronic documents;
retrieve a watched question from a watched question database, wherein the watched question is a question previously processed by a question answering pipeline of the data processing system and for which monitoring is performed to identify changes to previously generated answers for the watched question;
determine whether full processing of the watched question through the question answering pipeline is to be bypassed when performing watched question processing in response to the detected change in content of the electronic documents in the corpus;
in response to determining that full processing of the watched question through the question answering pipeline is to be bypassed:
bypass one or more stages of processing of the question answering pipeline; and
return a result of the watched question processing indicating that no changes to the previously generated answers are caused by the change in the content of the electronic documents in the corpus; and

in response to determining that full processing of the watched question through the question answering pipeline is not to be bypassed:
perform full processing of the watched question through all stages of the question answering pipeline to generate one or more new answers to the watched question based on the updated corpus;

wherein determining whether full processing of the watched question through the question answering pipeline is to be bypassed comprises:
obtaining first results comprising a first predetermined number of highest ranked documents returned by a primary search of the corpus prior to the change in content of electronic documents in the corpus;
obtaining second results comprising a second predetermined number of highest ranked documents returned by a primary search of the updated corpus after the change in content of electronic documents in the corpus;
comparing the first results to the second results; and
determining whether full processing of the watched qu

Assignees

• IBM

Inventors

• Boyer John M
• Kyle Bryan J

Classifications

• G06F16/00

  Information retrieval; Database structures therefor; File system structures therefor · CPC title

• G06N5/04 Primary

  Inference or reasoning models · CPC title