NATURAL LANGUAGE SEARCH RESULTS FOR INTENT QUERIES

• US 20160357860A1
• Filed: 08/16/2016
• Published: 12/08/2016
• Est. Priority Date: 06/04/2013
• Status: Abandoned Application

• Current Assignee: Google LLC (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: SHMIEL, Tomer, KEYSAR, Dvir, EREZ, Yonatan
• Application Number:
  US15/237,865
• Publication Number:
  US 20160357860A1
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/3329 Natural language query formulation or dialogue systems
• CPC Class Codes: G06F 16/334 Query execution G06F16/335 takes precedence
• CPC Class Codes: G06F 16/35 Clustering; Classification
• CPC Class Codes: G06F 16/93 Document management systems
• CPC Class Codes: G06F 40/186 Templates
• CPC Class Codes: G06F 40/205 Parsing
• CPC Class Codes: G06F 40/258 Heading extraction; Automatic titling; Numbering

Abstract

Systems and methods provide natural language search results to clear-intent queries. To provide the natural language search results, a system may parse a document from an authoritative source to generate at least one heading-text pair, the text appearing under the heading in the document. The system may assign a topic and a question category to the heading-text pair and store the heading-text pair in a data store keyed by the topic and the question category. The system determines that a query corresponds to the topic and the question category, and provides the heading-text pair as a natural language search result for the query. In some implementations, the text portion of the heading-text pair may be a paragraph or a list of items and the natural language search result may be provided with conventional snippet-based search results in response to the query.

Claims

1. A computer system comprising:

at least one processor; and

memory storing instructions that, when executed by the at least one processor, cause the computer system to perform operations comprising;

parsing documents from authoritative sources to generate a plurality of heading-text pairs, a heading-text pair being taken from document contents,

generating a set of potential templates from the heading-text pairs,

determining a quantity of occurrences for at least some of the set of potential templates,

storing potential templates with highest quantities as intent templates in a memory of the computer system; and

using the templates to populate a data store of heading-text pairs used to respond to natural language queries.

2. The computer system of claim 1, wherein converting the heading to potential templates includes replacing subsets of consecutive terms in the heading with a variable portion.

3. The computer system of claim 1, wherein the set of potential templates is a first set of potential templates and the memory stores instructions that, when executed by the at least one processor, cause the computer system to further perform operations including:

determining, using search records, previously issued queries that have search results associated with the authoritative sources;

generating a second set of potential templates from the determined previously issued queries; and

including the second set of potential templates with the first set of potential templates when determining the quantity of occurrences.

4. The computer system of claim 1, wherein the memory stores instructions that, when executed by the at least one processor, cause the computer system to further perform operations including:

assigning a question category to the intent templates, the question category being stored as an attribute of the intent template.

5. The computer system of claim 4, wherein the memory stores instructions that, when executed by the at least one processor, cause the computer system to further perform operations including:

receiving a natural language query;

determining an intent template of the intent templates that corresponds to the natural language query, the intent template having an associated question category;

determining a topic for the natural language query using the intent template;

searching an index of documents for documents responsive to the topic and the associated question category; and

providing a search result to for the natural language query that includes the documents responsive to the topic and the associated question category.

6. A system comprising:

at least one processor;

a data store of heading-text pairs keyed by topic and question category, the heading-text pairs being extracted from content of authoritative sources; and

memory storing instructions that, when executed by the at least one processor, cause the system to perform operations including;

determining that a query corresponds to an intent template of a plurality of intent templates, the intent template having an associated question category,

determining a topic for the query based on the intent template,

retrieving heading-text pairs from the data store that have a respective topic and question category key that corresponds with the topic for the query and the question category of the template, and

providing a search result for the query, wherein the search result includes at least one of the retrieved heading-text pairs.

7. The system of claim 6, wherein the intent template includes one non-variable portion and one variable portion, and wherein corresponding the query to the intent template includes:

determining that the query includes a first term that corresponds to the one non-variable portion;

determining that a second term in the query aligns with the variable portion; and

determining that the second term in the query corresponds to a topic in the data store.

8. The system of claim 7, wherein corresponding the query to the intent template includes:

generating potential templates from terms of the query; and

determining whether one of the potential templates corresponds to the intent template.

9. The system of claim 6, the memory storing instructions that, when executed by the at least one processor, cause the system to perform further operations including:

searching a document corpus using the topic for the query and the question category of the template as a new query instead of searching the document corpus with the query; and

providing search results from the document corpus with the at least one of the retrieved heading-text pairs.

10. A method comprising:

parsing a document from an authoritative source to generate at least one heading-text pair, the text appearing under the heading in the document;

assigning a topic and a question category to the heading-text pair;

storing the heading-text pair in a data store keyed by the topic and the question category;

determining that a query corresponds to the topic and the question category; and

providing the heading-text pair as a natural language search result for the query.

11. The method of claim 10, further comprising:

retrieve a plurality of heading-text pairs from the data store, each heading-text pair being keyed by the topic and the question category; and

rank the plurality of heading-text pairs; and

select a predetermined number of highest ranked heading-text pairs for the search result.

12. The method of claim 10, further comprising:

generate snippet-based search results by searching an index of documents for documents responsive to the query, and

provide the snippet-based search results with the natural language search result.

13. The method of claim 10, wherein assigning a topic to the heading-text pair includes:

identifying a topic for another heading text pair stored in the data store that matches a portion of a URL for the document and has the same question category as the heading-text pair; and

associating the heading-text pair with the topic of the other heading-text pair.

14. The method of claim 10, wherein assigning a topic to the heading-text pair includes:

identifying a topic for another heading-text pair stored in the data store that corresponds with dominant terms in the document and has the same question category as the heading-text pair; and

associating the heading-text pair with the topic of the other heading-text pair.

15. The method of claim 10, wherein determining that the query corresponds to the topic and question category includes:

determining that the query includes a topic used to key heading-text pairs in the data store; and

for each heading-text pair keyed by the topic in the data store;

issue a second query that includes the topic and question category for the heading-text pair,

compare search results for the second query to search results obtained from a document corpus for the query, and

determine that the query corresponds to the topic and question category of the heading-text pair when search results for the second query meet a similarity threshold with regard to the search results for the query from the document corpus.

16. The method of claim 10, further comprising memory storing a plurality of intent templates and wherein the heading-text pair is generated when the heading conforms to one of the plurality of intent templates.

17. The method of claim 16, wherein the question category for the heading-text pair is determined by the intent template the heading conforms to.

18. The method of claim 10, wherein generating the heading-text pair includes:

determining a topic from a context of the heading in the document; and

adding the topic to a heading portion of the heading-text pair.