Disambiguating organization names

• US 9,779,388 B1
• Filed: 12/09/2014
• Issued: 10/03/2017
• Est. Priority Date: 12/09/2014
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: LinkedIn Corporation (Microsoft Corporation)
• Inventors: Hansen, Axel R., Varon, Jonah L., Sinnott-Armstrong, Nicholas A., Hill, Shane S.
• Primary Examiner(s): Gonzales, Vincent
• Application Number:
  US14/565,165
• Time in Patent Office: 1,029 Days
• CPC Class Codes: G06Q 10/10 Office automation; Time management
• CPC Class Codes: G06Q 50/01 Social networking

Abstract

A system, method, and apparatus are provided for disambiguating organization names. Selected names that are shared among multiple organizations may or may not be categorized or characterized (e.g., by industry, by size, by reach). As content items are received (e.g., news stories, magazine articles, social media content), occurrences of the selected names are identified. Each item that includes at least one name is processed to determine which of the multiple entities that have the name (if any) is the organization referenced or mentioned in the item. The same model may be applied to disambiguate all names or, depending on the name'"'"'s categorization, different models or procedures may be applied to disambiguate the name.

Claims

1. A computer-implemented method of disambiguating organization names, the method comprising:

selecting multiple names of organizations; and

using the computer to;

extract a first name identified in an electronic content item, wherein multiple different organizations have the first name;

for each of the multiple organizations, assemble a corresponding feature vector comprising;

a set of name features reflecting use of the first name in the electronic content item; and

a set of affiliation features reflecting one or more affiliations of the organization also identified in the electronic content item;

execute a model to identify, among the multiple organizations, only one organization to which the electronic content item refers; and

make the electronic content item available to a user interested in content items that reference the one organization.

2. The method of claim 1, wherein executing a model comprises, for each organization in the multiple organizations:

applying the model to the organization'"'"'s corresponding feature vector; and

receiving as output from the model an indication as to whether the first name refers to the organization.

3. The method of claim 2, wherein the set of affiliation features comprises:

for each of one or more types of affiliations, an indication whether the electronic content item references the affiliation; and

an indication whether the electronic content item references a leader of the organization.

4. The method of claim 2, wherein the set of name features comprises:

a first alias feature indicating whether any alias forms of the name of the organization appear in the electronic content item; and

a second alias feature indicating a number of alias forms of the name of the organization that appear in the electronic content item.

5. The method of claim 4, wherein the set of name features further comprises:

if the organization is a commercial business, an indication whether a stock symbol of the organization appears in the electronic content item; and

a cosine similarity calculated between a theme of the electronic content item and a description of the organization at the professional social networking site.

6. The method of claim 1, wherein each organization'"'"'s corresponding feature vector further comprises:

a set of wrongness features derived from correlation of the electronic content item with one or more other content items known to not reference the organization.

7. The method of claim 1, further comprising categorizing the multiple names by, for each name:

identifying a total number of references to the name among multiple content items received during a threshold period of time, other than the electronic content item; and

if the total number of references exceeds a threshold quantity, categorizing the name as famous.

8. The method of claim 1, further comprising categorizing the multiple names by, for each name:

determining the number of organizations in the multiple organizations;

categorizing the name as common if the number of organizations is greater than a threshold; and

categorizing the name as uncommon if the number of organizations is less than the threshold.

9. The method of claim 1, further comprising:

categorizing each of the multiple names;

wherein executing a model comprises executing a model associated with a category assigned to the first name.

10. An apparatus for disambiguating organization names, the apparatus comprising:

one or more processors; and

logic comprising instructions that, when executed by the one or more processors, cause the apparatus to;

select multiple names of organizations;

extract a first name identified in an electronic content item, wherein multiple different organizations have the first name;

for each of the multiple organizations, assemble a corresponding feature vector comprising;

a set of name features reflecting use of the first name in the electronic content item; and

a set of affiliation features reflecting one or more affiliations of the organization also identified in the electronic content item;

execute a model to identify, among the multiple organizations, only one organization to which the electronic content item refers; and

make the electronic content item available to a user interested in content items that reference the one organization.

11. The apparatus of claim 10, wherein executing a model comprises, for each organization in the multiple organizations:

applying the model to the organization'"'"'s corresponding feature vector; and

receiving as output from the model an indication as to whether the first name refers to the organization.

12. The apparatus of claim 11, wherein the set of affiliation features comprises:

for each of one or more types of affiliations, an indication whether the electronic content item references the affiliation; and

an indication whether the electronic content item references a leader of the organization.

13. The apparatus of claim 11, wherein the set of name features comprises:

a first alias feature indicating whether any alias forms of the name of the organization appear in the electronic content item; and

a second alias feature indicating a number of alias forms of the name of the organization that appear in the electronic content item.

14. The apparatus of claim 13, wherein the set of name features further comprises:

if the organization is a commercial business, an indication whether a stock symbol of the organization appears in the electronic content item; and

a cosine similarity calculated between a theme of the electronic content item and a description of the organization at the professional social networking site.

15. The apparatus of claim 10, wherein each organization'"'"'s corresponding feature vector further comprises:

a set of wrongness features derived from correlation of the electronic content item with one or more other content items known to not reference the organization.

16. The apparatus of claim 10, wherein the logic further comprises instructions that, when executed by the one or more processors, cause the apparatus to categorize the multiple names by, for each name:

identifying a total number of references to the name among multiple content items received during a threshold period of time, other than the electronic content item; and

if the total number of references exceeds a threshold quantity, categorizing the name as famous.

17. The apparatus of claim 10, wherein the logic further comprises instructions that, when executed by the one or more processors, cause the apparatus to categorize the multiple names by, for each name:

determining the number of organizations in the multiple organizations;

categorizing the name as common if the number of organizations is greater than a threshold; and

categorizing the name as uncommon if the number of organizations is less than the threshold.

18. The apparatus of claim 10, wherein the logic further comprises instructions that, when executed by the one or more processors, cause the apparatus to:

categorize each of the multiple names;

wherein executing a model comprises executing a model associated with a category assigned to the first name.

19. A system for disambiguating organization names, comprising:

a physical storage device storing multiple names of organizations;

one or more processors;

logic comprising instructions that, when executed by the one or more processors, cause the system to;

select multiple names of organizations;

extract a first name identified in an electronic content item, wherein multiple different organizations have the first name;

for each of the multiple organizations, assemble a corresponding feature vector comprising;

a set of name features reflecting use of the first name in the electronic content item; and

a set of affiliation features reflecting one or more affiliations of the organization also identified in the electronic content item;

execute a model to identify, among the multiple organizations, only one organization to which the electronic content item refers; and

make the electronic content item available to a user interested in content items that reference the one organization.

20. The system of claim 19, further comprising a professional social networking site.