Identifying a stale data source to improve NLP accuracy

• US 10,387,468 B2
• Filed: 03/14/2013
• Issued: 08/20/2019
• Est. Priority Date: 03/12/2013
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Clark, Adam T., Dubbels, Joel C., Huebert, Jeffrey K., Petri, John E.
• Primary Examiner(s): Neway, Samuel G
• Application Number:
  US13/804,734
• Publication Number:
  US 20140280203A1
• Time in Patent Office: 2,350 Days
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

In some NLP systems, queries are compared to different data sources stored in a corpus to provide an answer to the query. However, the best data sources for answering the query may not currently be contained within the corpus or the data sources in the corpus may contain stale data that provides an inaccurate answer. When receiving a query, the NLP system may evaluate the query to identify a data source that is likely to contain an answer to the query. If the data source is not currently contained within the corpus, the NLP system may ingest the data source. If the data source is already within the corpus, however, the NLP may determine a time-sensitivity value associated with at least some portion of the query. This value may then be used to determine whether the data source should be re-ingested—e.g., the information contained in the corpus is stale.

Claims

1. A method comprising:

receiving a query for processing by a natural language processing (NLP) system comprising a corpus containing data ingested from a plurality of data sources, wherein the data is formatted and stored into one or more objects and organized based on topic changes;

identifying a data source expected to contain an answer to the query using NLP, by;

dividing words in the query into different elements, generating an annotation for each of the elements using the NLP system by determining a particular topic describing each of the elements, and

identifying a previously ingested data source in the corpus that is associated with previously-generated annotations matching the generated annotations for the elements;

upon determining that the previously ingested data in the corpus does not contain the answer to the query, determining whether new material has been added to the identified data source since the last time the identified data source was ingested into the corpus;

upon determining that new material has been added to the identified data source since the last time the identified data source was ingested into the corpus;

re-ingesting the identified data source using one or more computer processors whereby the new material is inserted into the corpus; and

processing the query to determine a lexical answer type for the query, based at least in part on a concept assigned to each of the elements, wherein the concepts were determined and assigned using NLP, and wherein the lexical answer type is a word or noun phrase that predicts a type of an answer to the query; and

generating an answer to the query based on the new material inserted into the corpus and based on the lexical answer type.

2. The method of claim 1, wherein determining that the ingested data in the corpus does not contain the answer to the query further comprises:

processing the query to determine if any of the data sources included in the corpus contain the answer to the query.

3. The method of claim 2, wherein determining whether the new material has been added to the identified data source since the last time the identified data source was ingested into the corpus is performed by the NLP system only if no answer is found when processing the query to determine if any of the data sources included in the corpus contain the answer to the query.

4. The method of claim 2, wherein determining that the ingested data in the corpus does not contain the answer to the query further comprises:

upon determining that the identified data source contains a possible answer to the query, determining a confidence score associated with the possible answer; and

upon determining that the confidence score does not satisfy a confidence threshold, determining whether the new material has been added to the identified data source since the last time the identified data source was ingested into the corpus.

5. The method of claim 1, wherein identifying the data source expected to contain the answer to the query further comprises:

performing a concept mapping to assign a concept to at least one element in the query; and

determining that the assigned concept of the at least one element in the query is related to a concept associated with the ingested data in the corpus from the data source.

6. The method of claim 5, wherein the corpus stores metadata describing each of the plurality of data sources, wherein the assigned concept of the at least one element in the query is related to the metadata description of the identified data source.

7. The method of claim 1, wherein the corpus comprises a plurality of data from different data sources, wherein the data from the different data sources is organized based on a common format of the corpus.

8. The method of claim 1, wherein the one or more features indicative of a topic change are one or more changes in text formatting in the corpus.