Granted Patent B1

US 10,963,501 · App. 15/582,625 · Granted Mar 30, 2021

Inventors: Naveen Ramachandrappa (San Jose, CA); Ramya Mula (San Jose, CA); Ashwin Kayyoor (Sunnyvale, CA); Bashyam Tca (Saratoga, CA)

Assignee: Veritas Technologies LLC

G06F16/3347G06F7/08G06F16/322

Abstract

Claims (44)

1. A computer-implemented method for generating a topic tree for digital information to classify said digital information into a hierarchy of topics and subtopics, the method comprising:

parsing the digital information to locate at least one sentence;

determining at least one noun in the digital information;

generating at least one word that is similar to the at least one noun;

removing at least one proper noun from the digital information, wherein the proper noun includes at least one of: a person's name, a location's name, or an organization's name;

extracting a set of keywords from the digital information, wherein the set of keywords comprises the at least one noun and the at least one word that is similar to the at least one noun;

comparing the set of keywords to an ontology;

extracting at least one hierarchy from the ontology that matches the set of keywords by comparing the set of keywords to a plurality of hierarchies in the ontology, selecting all hierarchies that include at least one keyword from the set of keywords, and removing all hierarchies from the selected hierarchies that do not include a threshold number of keywords;

generating word embeddings that encode information about a correlation with other keywords using cosine similarity between vectors to find correlated keywords between the set of keywords;

merging all the hierarchies extracted from the ontology; and

sorting the merged hierarchies extracted from the ontology based on relevance scores from the word embeddings for the topic tree for digital information.

2. The method according to claim 1 , further comprising mapping the digital information to weighted vectors, wherein the sorting of the at least one extracted hierarchy is based on the weighted vectors.

3. The method according to claim 2 , further comprising generating a set of similar words based on a cosine similarity between the weighted vectors.

4. The method according to claim 1 , wherein extracting the set of keywords comprises:

locating a plurality of sentences within the digital information; and

applying part-of-speech tagging to the located plurality of sentences.

5. A system for generating a topic tree for digital information to classify said digital information into a hierarchy of topics and subtopics, the system comprising:

a parsing module, stored in memory, that parses the digital information to locate at least one sentence, determines at least one noun in the digital information, generates at least one word that is similar to the at least one noun, removes at least one proper noun from the digital information, wherein the proper noun includes at least one of: a person's name, a location's name, or an organization's name, and extracts a set of keywords, wherein the set of keywords comprises the at least one noun and the at least one word that is similar to the at least one noun;

a comparison module, stored in memory, that compares the set of keywords to an ontology; an extraction module, stored in memory, that extracts a plurality of hierarchies from the ontology that match the set of keywords by comparing the set of keywords to a plurality of hierarchies in the ontology, selects all hierarchies that include at least one keyword that match from the set of keywords, removes all hierarchies from the selected hierarchies that do not include a threshold number of keywords, generates word embeddings that encode information about a correlation with other keywords using cosine similarity between vectors to find correlated keywords between the set of keywords, and merges all the hierarchies extracted from the ontology;

a sorting module, stored in memory, that sorts the merged hierarchies extracted from the ontology based on relevance scores from the word embeddings for the topic tree for digital information; and

at least one processor that executes the parsing module, the comparison module, the extraction module, and the sorting module.

6. The system according to claim 5 , further comprising a mapping module, stored in memory, that maps the digital information to weighted vectors, wherein the sorting module sorts the extracted hierarchies based on the weighted vectors.

7. The system according to claim 6 , wherein the mapping module generates a set of similar words based on a cosine similarity between the weighted vectors.

8. The system according to claim 5 , wherein the parsing module extracting the set of keywords:

locates a plurality of sentences within the digital information; and

applies part-of-speech tagging to the located plurality of sentences.

9. A non-transitory computer-readable medium comprising one or more computer-executable instructions that, when executed by at least one processor of a computing device generate a topic tree for digital information to classify said digital information into a hierarchy of topics and subtopics by causing the computing device to:

parse a digital document to locate at least one sentence;

determine at least one noun in the digital document;

generate at least one word that is similar to the at least one noun;

remove at least one proper noun from the digital document, wherein the proper noun includes at least one of: a person's name, a location's name, or an organization's name;

extract a set of keywords from the digital document, wherein the set of keywords comprises the at least one noun and the at least one word that is similar to the at least one noun;

compare the set of keywords to an ontology;

extract at least one hierarchy from the ontology that matches at least one keyword of the set of keywords by comparing the set of keywords to a plurality of hierarchies in the ontology, selecting all hierarchies that include at least one keyword from the set of keywords, and removing all hierarchies from the selected hierarchies that do not include a threshold number of keywords;

generate word embeddings that encode information about a correlation with other keywords using cosine similarity between vectors to find correlated keywords between the set of keywords;

merge all the hierarchies extracted from the ontology; and

sort the merged hierarchies from the ontology based on relevance scores from the word embeddings for the topic tree for digital information.

10. The non-transitory computer-readable medium according to claim 9 , wherein:

the one or more computer-executable instructions cause the computing device to map the digital information to weighted vectors; and

the sorting of the at least one extracted hierarchy is based on the weighted vectors.

11. The non-transitory computer-readable medium according to claim 10 , wherein the one or more computer-executable instructions cause the computing device to generate a set of words that are similar to the keywords based on a cosine similarity between the weighted vectors.

12. The non-transitory computer-readable medium according to claim 9 , wherein extracting the set of keywords comprises:

locating a plurality of sentences within the digital information; and

applying a part-of-speech tagging to the located plurality of sentences.

Assignments (14)

Cited By (3)

US 12,204,860 US 12,222,974 US 12,242,803