Dynamic faceted search on a document corpus

Patent metadata

Field	Value
Publication number	US-11275796-B2
Application number	US-201916399030-A
Country	US
Kind code	B2
Filing date	Apr 30, 2019
Priority date	Apr 30, 2019
Publication date	Mar 15, 2022
Grant date	Mar 15, 2022

Abstract

A query-focused faceted structure generation method, system, and computer program product for generating a query-focused faceted structure from a taxonomy for searching a document collection, including ingesting a document corpus, generating a vector space representation of a query and instances from a taxonomy of the document corpus, and producing a dynamic structure of a relevant facet categories and facet values using a two-vector space representation from the generated vector space representation.

First claim

What is claimed is:

1. A computer-implemented query-focused faceted structure generation method for generating a query-focused faceted structure from a taxonomy for searching a document collection, the method comprising:
ingesting a document corpus including a pre-processing that filters parts of speech;
generating a vector space representation of a query and instances from a taxonomy of the document corpus via at least two models, the taxonomy being loaded and including a graph of a type and instance nodes where the instance nodes have a consistent relationship to the type; and
producing a dynamic structure of a relevant category and facet using a two-vector space representation from the generated vector space representation based on a separate two-vector space representation of the at least two models,
wherein the ingesting ingests the document corpus by:
extracting the terminology that includes noun words and phrases from the document corpus to:
train a type model that generates a phrase embedding of the terminology in the document corpus; and
train a topic model that generates a second phrase embedding of the terminology in the document corpus,

wherein the generating generates a vector for a user query as a weighted combination of the vector for each query token in the topic model as a query vector,
wherein the generating generates a list of the vectors for instances from the taxonomy in the topic model, and
wherein the producing produces the dynamic structure of the relevant category and the facet by:
selecting a first parameter of nearest neighbor instances to the query vector from the taxonomy instances using the topic model as query-similar instances;
selecting a second parameter of types in the taxonomy with a most number of query-similar instances to use as categories;
selecting a third parameter of facets from instances of the types corresponding to each of the categories for the second parameter; and
expanding from the third parameter of the facets within each of the second parameter of the categories to obtain more category-similar instances from the document corpus using the type model.

2. The method of claim 1 , further comprising returning the dynamic structure as a data file to a user.

3. The method of claim 1 , wherein the facets are ranked within each of the first parameter of the categories by distance to both:
the query vector in the topic model vector space, and
a centroid of the third parameter of instances that correspond to the category.

4. The method of claim 1 , embodied in a cloud-computing environment.

5. A computer program product for query-focused faceted structure generation, the computer program product comprising a computer-readable storage medium having program instructions embodied therewith for generating a query-focused faceted structure from a taxonomy for searching a document collection, the program instructions executable by a computer to cause the computer to perform:
ingesting a document corpus including a pre-processing that filters parts of speech;
generating a vector space representation of a query and instances from a taxonomy of the document corpus via at least two models, the taxonomy being loaded and including a graph of a type and instance nodes where the instance nodes have a consistent relationship to the type; and
producing a dynamic structure of a relevant category and facet using a two-vector space representation from the generated vector space representation based on a separate two-vector space representation of the at least two models,
wherein the ingesting ingests the document corpus by:
extracting the terminology that includes noun words and phrases from the document corpus to:
train a type model that generates a phrase embedding of the terminology in the document corpus; and
train a topic model that generates a second phrase embedding of the terminology in the document corpus,

wherein the generating generates a vector for a user query as a weighted combination of the vector for each query token in the topic model as a query vector,
wherein the generating generates a list of the vectors for instances from the taxonomy in the topic model, and
wherein the producing produces the dynamic structure of the relevant category and the facet by:
selecting a first parameter of nearest neighbor instances to the query vector from the taxonomy instances using the topic model as query-similar instances;
selecting a second parameter of types in the taxonomy with a most number of query-similar instances to use as categories;
selecting a third parameter of facets from instances of the types corresponding to each of the categories for the second parameter; and
expanding from the third parameter of the facets within each of the second parameter of the categories to obtain more category-similar instances from the document corpus using the type model.

6. The computer program product of claim 5 , further comprising returning the dynamic structure as a data file to a user.

7. The computer program product of claim 5 , wherein the facets are ranked within each of the first parameter of the categories by distance to both:
the query vector in the topic model vector space, and
a centroid of the third parameter of instances that correspond to the category.

8. A query-focused faceted structure generation system for generating a query-focused faceted structure from a taxonomy for searching a document collection, the system comprising:
a processor; and
a memory, the memory storing instructions to cause the processor to perform:
ingesting a document corpus including a pre-processing that filters parts of speech;
generating a vector space representation of a query and instances from a taxonomy of the document corpus via at least two models, the taxonomy being loaded and including a graph of a type and instance nodes where the instance nodes have a consistent relationship to the type; and

producing a dynamic structure of a relevant category and facet using a two-vector space representation from the generated vector space representation based on a separate two-vector space representation of the at least two models,
wherein the ingesting ingests the document corpus by:
extracting the terminology that includes noun words and phrases from the document corpus to:
train a type model that generates a phrase embedding of the terminology in the document corpus; and
train a topic model that generates a second phrase embedding of the terminology in the document corpus,

wherein the generating generates a vector for a user query as a weighted combination of the vector for each query token in the topic model as a query vector,
wherein the generating generates a list of the vectors for instances from the taxonomy in the topic model, and
wherein the producing produces the dynamic structure of the relevant category and the facet by:
selecting a first parameter of nearest neighbor instances to the query vector from the taxonomy instances using the topic model as query-similar instances;
selecting a second parameter of types in the taxonomy with a most number of query-similar instances to use as categories;
selecting a third parameter of facets from instances of the types corresponding to each of the categories for the second parameter; and
expanding from the third parameter of the facets within each of the second parameter of the categories to obtain more category-similar instances from the document corpus using the type model.

9. The system of claim 8 , further comprising returning the dynamic structure as a data file to a user.

10. The system of claim 8 , embodied in a cloud-computing environment.

Assignees

• IBM

Inventors

• Kong Biying
• Rajshree Nidhi
• Gliozzo Alfio Massimiliano
• Fauceglia Nicolas Rodolfo
• Farrell Robert G
• Chowdhury Md Faisal Mahbub
• Mathur Anish

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06N20/00

  Machine learning · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title

• G06F16/93 Primary

  Document management systems · CPC title