Granted Patent B2

US 7,657,546 · App. 11/340,246 · Granted Feb 2, 2010

Assignee: International Business Machines Corporation

Abstract

Claims (54)

1. A method of automatically discovering ontology file categories, said method comprising the steps of:

a) searching for available semantic data files;

b) storing links and content to identified semantic data files;

c) generating an ontology file from stored said content for each linked said semantic data file;

d) identifying a domain for each said ontology file, said domain being identified from generated ontology files;

e) extracting a plurality of ontology file categories from domains identified for said generated ontology files, said ontology file categories being statistically identified automatically from said domains, extracting comprising:

determining and normalizing contextual significance for all domains, each normalized contextual significance providing a significance value for a respective domain, and

combining discovered domains and features for generated ontology files responsive to domain significance values;

f) providing a training set from generated ontology files, said training set including an instance set, a domain set and a feature set; and

g) classifying ontology file instances responsive to said training sets, results of classification indicating automatic category discovery effectiveness.

2. A method as in claim 1 , the step (a) of searching comprising the steps of:

i) searching the Internet for semantic web pages; and

ii) storing links to identified said semantic web pages in said link database.

3. A method as in claim 2 , wherein the step (i) of searching the Internet comprises limiting a network crawler to locating only web pages available over the Internet that are written in a semantic markup language, said method further comprising:

h) classifying ontology file instances from one or more groups of ontology files responsive to said classification results to refine ontology file categories;

j) receiving new/modified ontology files; and

k) returning to step (d) to identify domains for said new/modified ontology files and update automatically discovered said ontology file categories.

4. A method as in claim 1 , the step (b) of storing links comprising the steps of:

i) storing said links in a link database;

ii) applying categories to linked said semantic data files; and

iii) downloading content from categorized said linked semantic data files, downloaded said content being stored in a content database.

5. A method as in claim 4 , wherein the step (ii) of applying categories comprises grouping links stored in said link database by attribute.

6. A method as in claim 5 , wherein said links are grouped by Internet domain and language; and, time and size.

7. A method as in claim 1 , the step (c) of generating an ontology file comprising the steps of:

i) parsing said content;

ii) extracting names from parsed said content; and

iii) converting said content into ontology files responsive to extracted said names and guided by a natural language processor filtering, cleaning and segmenting said names and identifying synonyms, acronyms and antonyms for said names.

8. A method as in claim 7 , wherein the step (ii) of extracting extracts names from parsed said content according to class, property and instance for text mining.

9. A method as in claim 1 , the step (d) of identifying domains comprising the steps of:

i) selecting keywords from said each ontology file;

ii) filtering a sense from selected said keywords responsive to a lexical database; and

iii) identifying a domain in said each ontology file from said selected keywords.

10. A method as in claim 9 , wherein the step (ii) of filtering senses filters synsets for each keyword.

11. A method as in claim 1 , the step (g) classifying ontology file instances comprising the steps of:

i) providing classification parameters from said training sets to a classification unit; and

ii) classifying ontology instances responsive to said classification parameters, said classification unit receiving updated classification parameters with each change in ontology files and reclassifying said ontology instances with said each change.

12. A method of automatically discovering ontology file categories, said method comprising the steps of:

a) searching for available semantic data files;

b) storing links and content to identified semantic data files;

c) generating an ontology file from stored said content for each linked said semantic data file;

d) identifying a domain for each said ontology file, said domain being identified from generated ontology files, identifying domains comprising the steps of:

i) selecting keywords from said each ontology file,

ii) filtering a sense from selected said keywords responsive to a lexical database, wherein filtering senses filters synsets for each keyword, and

iii) identifying a domain in said each ontology file from said selected keywords;

e) extracting a plurality of ontology file categories from domains identified for said generated ontology files, said ontology file categories being statistically identified automatically from said domains, wherein extracting categories comprises the steps of:

A) measuring sense significance from filtered said synsets and providing a context measure of said each ontology file,

B) defining a feature set containing significant senses for said each ontology file,

C) perusing the said filtered synsets and selecting one sense for said each ontology file, said one sense being a domain representing said each ontology file,

D) normalizing contextual significance for all domains, each normalized contextual significance providing a significance value for a respective domain, and

E) combining discovered domains and features for generated ontology files responsive to domain significance values;

f) providing a training set from generated ontology files, said training set including an instance set, a domain set and a feature set; and

g) classifying ontology file instances responsive to said training sets, results of classification indicating automatic category discovery effectiveness.

13. A method as in claim 12 , wherein the step (B) of defining said feature set comprises identifying significant synsets of said each ontology file, each synset representing a domain in an ontology file.

14. A method as in claim 13 , wherein the step (C) of identifying said domain further comprises collecting statistics of discovered domains, collected said statistics selecting said categories representing said generated ontology files, collected said statistics providing metrics for measuring the effectiveness of category discovery.

Assignments (8)

Continuity (1)