Automatic disambiguation based on a reference resource

• US 8,112,402 B2
• Filed: 02/26/2007
• Issued: 02/07/2012
• Est. Priority Date: 02/26/2007
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Cucerzan, Silviu-Petru, Schultz, Mike
• Primary Examiner(s): Ali, Mohammad
• Assistant Examiner(s): Corbo, Griselle
• Application Number:
  US11/710,805
• Publication Number:
  US 20080208864A1
• Time in Patent Office: 1,807 Days
• Field of Search: 707/6, 707/736, 707748-750, 707/758, 707/705
• US Class Current: 707/705
• CPC Class Codes: G06F 40/295 Named entity recognition

Abstract

A novel system for automatically indicating the specific identity of ambiguous named entities is provided. An automatic disambiguation data collection is created using a reference resource. Explicit named entities are catalogued from the reference resource, together with various abbreviated, alternative, and casual ways of referring to the named entities. Entity indicators, such as labels and context indicators associated with the named entities in the reference resource, are also catalogued. The automatic disambiguation collection can then be used as a basis for evaluating ambiguous references to named entities in text content provided in different applications. The content surrounding the ambiguous reference may be compared with the entity indicators to find a good match, indicating that the named entity associated with the matching entity indicators is the intended identity of the ambiguous reference, which can be automatically provided to a user.

Claims

1. A computer implemented method, performed by a computer having a processor, of disambiguating references to named entities, comprising:

identifying a surface form of a named entity in a text, the surface form being an ambiguous orthographic representation of a common name for the named entity, the surface form having a corresponding surface form reference in a surface form reference database;

enumerating, from the surface form reference, a plurality of different reference named entities based on the identified surface form of the named entity, wherein the surface form is associated in the surface form reference with the plurality of different reference named entities each being formed of a different set of words, and each of the different reference named entities is associated with a named entity reference, the named entity references being stored in a named entity reference database that is separate from the surface form reference database, each of the named entity references associating one of the different reference named entities to multiple entity indicators, the entity indicators including both labels applied to a respective named entity in an information resource, and context indicators applied to the respective named entity in the information resource, in which the labels comprise classifying identifiers applied to the respective named entities in the information resource;

evaluating, with the processor, one or more measures of correlation between one or more of the entity indicators in the information resource for each of the identified reference named entities, and the text, the evaluation including comparisons of the text to both the labels and the context indicators;

identifying, with the processor, one of the reference named entities for which the associated entity indicators have a relatively high correlation to the text; and

providing a disambiguation output that indicates the identified reference named entity to be associated with the surface form of the named entity in the text.

2. The method of claim 1, wherein the labels comprise at least one element selected from among:

a tag;

a category;

a heading;

a title;

a section name; and

a list name.

3. The method of claim 1, wherein the context indicators comprise at least one element selected from among:

an attribute included in a title of a document about the respective named entity in the information resource;

an attribute appearing proximate to the respective named entity in the information resource;

an attribute that links to a document about the respective named entity in the information resource; and

an attribute that is linked from a document about the respective named entity in the information resource.

4. The method of claim 1, in which the information resource is among one or more information resources, the method further comprising:

extracting a set of surface forms and entity indicators associated with a plurality of named entities from the one or more information resources;

storing the surface forms and named entities in a surface form reference, comprising a data collection indexed by the surface forms and indicating the named entities associated with each of the surface forms; and

storing the named entities and entity indicators in a named entity reference, comprising a data collection indexed by the named entities and containing the entity indicators associated with each of the named entities;

in which the surface form of a named entity in a text is identified from among the surface forms stored in the surface form reference, and the one or more measures of correlation are evaluated between the text and the extracted entity indicators stored in the named entity reference.

5. The method of claim 4, wherein the one or more information resources are selected from among a group consisting of:

an encyclopedia;

a directory;

an atlas; and

a citation index.

6. The method of claim 4, wherein the surface forms of a respective named entity are extracted from at least one element from among a group consisting of:

a title of a document about the respective named entity;

a title of a document that redirects to the document about the respective named entity;

a title of a document that disambiguates a surface form to the respective named entity as one possible disambiguation; and

the text of a link to a document about the respective named entity.

7. The method of claim 4, wherein the extracted entity indicators comprise context indicators extracted from at least one of the information resources, wherein the context indicators comprise elements of text associated with the respective named entities, in at least one of the information resources.

8. The method of claim 4, wherein the extracted entity indicators comprise labels extracted from at least one of the information resources.

9. The method of claim 1, further comprising:

identifying a plurality of surface forms that overlap within the text;

identifying one of the overlapping surface forms that is associated with one of the reference named entities in the surface form reference, for which the associated entity indicators have a relatively high correlation to the text; and

providing a disambiguation output that indicates the identified surface form and the reference named entity associated with the identified surface form.

10. The method of claim 1, further comprising receiving a user input indicating the text comprising the surface form for which a corresponding reference named entity is to be identified.

11. The method of claim 1, further comprising receiving an indication, from a text-associated application, of a block of text associated with the application, and providing one or more disambiguation outputs that indicate identified reference named entities to be associated with one or more surface forms within the block of text.

12. The method of claim 11, in which the text-associated application comprises at least one application selected from among a group of options consisting of:

a word processing application, a presentation application, a spreadsheet application, a database application, an email application, an information manager application, a web page application, a web page editing application, a reference work application, a fixed-layout document application, a desktop publishing application, a diagramming application, a project management application, a form developer application, a customer relationship management application, an enterprise resource planning application, a field services application, a retail management application, a social networking application, a tax preparation application, a finance management application, an accounting application, and a game application.

13. The method of claim 1, wherein evaluating the one or more measures of similarity comprises applying different weights to at least one of:

different types of the entity indicators;

proximity of the entity indicators to the surface form of the named entity within the text;

number of common labels;

or number of links among each other of entity indicators extracted from documents that link to or are linked from a document about an associated reference named entity.

14. The method of claim 1, wherein evaluating the one or more measures of similarity comprises using at least one element from a group consisting of:

a vector space model;

Jensen-Shannon divergence;

Kullback-Liebler divergence; and

mutual information.

15. The method of claim 1, wherein the text to be evaluated for disambiguation of surface forms may be selected from among a group of optional units of text consisting of:

a document, a page, a paragraph, a sentence, and a manually selected portion of text.

16. The method of claim 15, further comprising first evaluating the one or more measures of similarity between a larger one of the units of text and the one or more entity indicators, and if more than one of the reference named entities is identified as having associated entity indicators with a relatively high correlation to the entity indicators in the text, then evaluating the one or more measures of correlation between an iteratively smaller one of the units of text and the one or more entity indicators, until a unique reference named entity is identified as having associated entity indicators with a relatively high correlation to the entity indicators in the text.

17. A computer-readable storage medium comprising computer-executable instructions which, when executed by a computing device having a processor, enable the computing device to prepare and apply an automatic disambiguation system, comprising steps of:

extracting, with the processor, a collection of surface forms associated with a plurality of named entities, that are different from the surface forms, from an information resource;

extracting, with the processor, a collection of labels associated with the named entities from the information resource;

extracting, with the processor, a collection of context indicators associated with the named entities from the information resource;

when provided with a surface form in a text sample having different units of text, evaluating, with the processor, a measure of correlation of entity indicators associated with the surface form in the text sample with the labels and the context indicators associated with the named entities associated with the surface form in the collection of surface forms by first evaluating one or more measures of similarity between a larger one of the units of text and the entity indicators, and if more than one of the collection of extracted named entities from the information resource is identified as having associated entity indicators with a relatively high correlation to the entity indicators for the surface form in the sample of text, then evaluating the one or more measures of similarity between an iteratively smaller one of the units of text and the entity indicators, until a unique extracted named entity from the information resource is identified as having associated entity indicators with a relatively high correlation to the entity indicators for the surface form in the sample of text; and

providing a display, based on the measure of correlation, showing a representation of the text sample, the display including an indication of one of the named entities to be a disambiguation of the surface form in the text sample, the indication of the one of the named entities being positioned proximate to the surface form in the representation of the text sample.