HIERARCHIC MODEL AND NATURAL LANGUAGE ANALYZER

• US 20180089242A1
• Filed: 09/26/2016
• Published: 03/29/2018
• Est. Priority Date: 09/26/2016
• Status: Active Grant

• Current Assignee: Scotilev Limited
• Original Assignee: Twiggle, Ltd.
• Inventors: LEV, Iddo, SHALOM, Eli, AVIDOR, Adi
• Granted Patent: US 10,067,965 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 16/212 with details for data modelling support
• CPC Class Codes: G06F 16/2228 Indexing structures
• CPC Class Codes: G06F 16/248 Presentation of query results
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

A system and method for interpretation of product-related data, including extracting product information from a product-related dataset of an e-commerce data source, determining at least one representation of the product information by elements of a product database model, by finding model elements corresponding to elements of the product information and selecting model elements that form consistent representation of the product information with hierarchic relations between the model elements, and creating an indexed product database by indexing according to the representing model elements the at least one determined representation of the product information.

Claims

1. A system for creating a response to a search query using product-related data, comprising:

at least one memory which stores a product database and a hierarchic product model arranging a plurality of product categories in a hierarchy, said hierarchic product model defines a plurality of attributes for each one of said plurality of product categories and a plurality of values for each of said plurality of attributes;

at least one hardware processor executing a code for;

receiving—

natural language product information for each of a plurality of products from a product-related dataset of an e-commerce data source, said natural language product information defining product elements of a respective said product;

for each of said plurality of products automatically indexing one of a plurality of modeled product representations into said product database according to said hierarchic product model, said indexing is performed by;

selecting model attributes from said plurality of attributes of said hierarchic product model, said selected attributes are corresponding to at least one of synonyms and similar terms of the product elements from a respective said natural language product information of the respective product;

extracting from the hierarchic product model a plurality of hierarchic relations between the selected model attributes; and

indexing a respective said modeled product representation of the respective product in said product database according to the selected model attributes and the plurality of hierarchic relations;

modeling an input text defining a natural language search query by using said hierarchic product model to create a modeled search query; and

automatically creating a response to said natural language search query by matching said modeled search query with some of said plurality of modeled product representations defined in said indexed product database.

2. The system of claim 1, wherein the processor is configured to execute code for detecting among the selected model attributes a model attribute of high hierarchy relative to the other selected model attributes, and filtering out possible representing model attributes which are inconsistent with the high hierarchy model attribute.

3. The system of claim 1, wherein each of the selected model attributes represents at least one of:

a list comprising a product-related concept, an attribute of a concept, an enriched attribute depending on other attributes, a value of an attribute or a limitation on an attribute'"'"'s value space;

wherein a concept is a model attribute representing a certain kind of products or parts of products and an attribute is a model attribute representing properties of the concept.

4. The system of claim 3, wherein the processor is configured to execute code for receiving along with natural language product information an indication about a product, a concept, an attribute and/or a value thereof, to which the natural language product information relate, and limiting the selected model attributes to the indicated product, concept, attribute and/or value.

5. The system of claim 3, wherein each of the selected model attributes includes an instance of a concept representing a concept with a specific value of at least one of its attributes.

6. The system of claim 5, wherein each of the selected model attributes has secondary attributes.

7. The system of claim 1, wherein the hierarchic relations include at least one of inheritance and partonomy.

8. The system of claim 1, wherein each of the selected model attributes corresponding to a term includes the same term, a similar term and/or a synonym of the term.

9. The system of claim 1, wherein the processor is configured to execute code for the modeling an input text by determining at least one representation of the natural language search query received from a search server by elements of the hierarchic product model which are corresponding to elements of the natural language search query;

wherein said selected model attributes form a consistent representation of the natural language search query with hierarchic relations between the model attributes;

wherein the processor is configured to execute code for the automatically creating a response by transmitting the determined at least one representation to the search server, thus enabling the search server to match the at least one representation of the natural language search query to modeled representations of products stored in the indexed product database.

10. The system of claim 9, wherein the processor is configured to execute code for filtering possible representing model attributes in a pre-defined manner when corresponding pre-defined expressions are included in the natural language search query, a pre-defined expression is indicative of at least one of a list comprising a relation, a threshold, and a value restriction.

11. The system of claim 10, wherein the pre-defined expressions include at least one of a list comprising the expressions:

less than, more than, between, cheap, large, small and best for.

12. The system of claim 10, wherein the processor is configured to execute code for filtering out all products with a price above a predetermined threshold value when a corresponding expression is included in the natural language search query.

13. The system of claim 12, wherein the threshold value depends on a concept identified in the natural language search query.

14. The system of claim 10, wherein the processor is configured to execute code for calculating, in response to an expression included in the natural language search query, a certain value of an enriched attribute based on one or more attributes, and filtering out products that don'"'"'t fit the value.

15. The system of claim 1, wherein the processor is configured to execute code for identifying a model attribute to which a value in the natural language product information refers, according to a space of possible values of the a model attributes implemented in the database model.

16. The system of claim 1, wherein the processor is configured to execute code for ranking a possible representing model attribute according to frequency of use deduced from value distributions implemented in the database model.

17. The system of claim 1, wherein the processor is configured to execute code for identifying a model attribute to which a value in the natural language product information refers, according to data implemented in the hierarchic product model ascribing customary attributes to corresponding concepts.

18. The system of claim 1, wherein the processor is configured to execute code for identifying to which of several possible model attributes the natural language product information refers, according to the attributes each of the possible model attributes can have.

19. The system of claim 1, wherein the processor is configured to execute code for interpreting separate terms of the natural language product information as a concept, an attribute of the concept and/or a value of the attribute, respectively.

20. The system of claim 1, wherein the processor is configured to execute code for calculating a Boolean value of a model attribute and filter out possible representations based on the calculated Boolean value.

21. A method for creating a response to a natural language search query using product-related data, comprising:

providing a product database and a hierarchic product model arranging a plurality of product categories in a hierarchy, said hierarchic product model defines a plurality of attributes for each one of said plurality of product categories and a plurality of values for each of said plurality of attributes;

receiving natural language product information for each of a plurality of products from a product-related dataset of an e-commerce data source;

for each of said plurality of products automatically indexing one of a plurality of modeled product representations into said product database according to said hierarchic product model, said indexing is performed by;

selecting model attributes from said plurality of attributes of said hierarchic product model, said selected attributes are corresponding to at least one of synonyms and similar terms of product elements from a respective said natural language product information of the respective product;

extracting from the hierarchic product a plurality of hierarchic relations between the selected model attributes; and

indexing a respective said modeled product representation of the respective product in said product database according to;

the selected model attributes and the plurality of hierarchic relations;

modeling an input text defining a natural language search query by using said hierarchic product model to create a modeled search query; and

automatically creating a response to said natural language search query by matching said modeled search query with some of said plurality of modeled product representations defined in said indexed product database.

22. The method of claim 21, wherein said modeling comprises:

determining at least one representation of the natural language search query received from a search server by elements of the hierarchic product model, by finding model attributes corresponding to elements of the natural language search query in the hierarchic product model and selecting model attributes that form consistent representation of the natural language search query with hierarchic relations between the model attributes;

wherein said automatically creating comprises transmitting the at least one representation to the search server, thus enabling the search server to match the at least one representation of the natural language search query to modeled representations of products stored in the indexed product database.