NAME RECOGNITION

• US 20150161519A1
• Filed: 05/01/2013
• Published: 06/11/2015
• Est. Priority Date: 04/19/2013
• Status: Abandoned Application

• Current Assignee: Google Inc. (Alphabet Inc.)
• Original Assignee: Google Inc. (Alphabet Inc.)
• Inventors: Horling, Bryan Christopher, Zhong, Hongtao
• Application Number:
  US13/874,717
• Publication Number:
  US 20150161519A1
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 40/295 Named entity recognition
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

A computer-implemented technique includes obtaining training electronic messages, identifying name context in the training electronic messages, and determining patterns from the name context. The technique can include applying the patterns to the training electronic messages to extract candidate names and selecting a set of the patterns based on the extracted candidate names to obtain a set of patterns. In some implementations, the technique can further include applying the set of patterns to electronic messages associated with a first user having a registered profile, extracting candidate names, and selecting a set of alternate names for the first user from the candidate names. The technique can also include detecting a use of one alternate name from the set of alternate names by a second user, and outputting a suggestion to the second user in response to the detecting, the suggestion being based on the registered profile of the first user.

Claims

1. A computer-implemented method, comprising:

obtaining, at a server including one or more processors, training electronic messages;

identifying, at the server, one or more name contexts in the training electronic messages;

determining, at the server, patterns from the name contexts, each pattern including a context around a name and an associated position for the name relative to the context;

applying, at the server, the patterns to the training electronic messages to extract candidate names that correspond to the associated positions to obtain extracted candidate names;

selecting, at the server, a set of the patterns based on the extracted candidate names; and

storing, at the server, the set of patterns.

2. The computer-implemented method of claim 1, wherein the training electronic messages are obtained from plurality of training users, and wherein each specific training electronic message includes at least one known name associated with a specific field of the specific training electronic message.

3. The computer-implemented method of claim 2, wherein identifying the one or more name contexts in the training electronic messages includes identifying, at the server, N tokens surrounding each known name, wherein each token is a word or a punctuation, and wherein N is an integer greater than zero.

4. The computer-implemented method of claim 3, wherein determining the patterns includes determining, at the server, context for each combination of the N tokens surrounding the known name and determining the associated position at the known name to obtain the patterns.

5. The computer-implemented method of claim 1, wherein selecting the set of the patterns includes selecting each pattern that, when applied to the training electronic messages, extracts candidate names having greater than a first predetermined matching accuracy with actual names in the training electronic messages.

6. The computer-implemented method of claim 1, further comprising:

obtaining, at the server, electronic messages associated with a first user, the first user having a registered profile;

applying, at the server, the set of patterns to the electronic messages to extract candidate names for the first user;

selecting, at the server, a set of the candidate names having greater than a predetermined usage rate in the electronic messages to obtain a set of alternate names for the first user; and

storing, at the server, the set of alternate names for the first user.

7. The computer-implemented method of claim 6, further comprising:

detecting, at the server, a use of one alternate name from the set of alternate names by a second user at a computing device; and

outputting, from the server, a suggestion for the second user to the computing device, the suggestion being based on the registered profile for the first user.

8. The computer-implemented method of claim 7, wherein outputting the suggestion causes the computing device to automatically select a name for the first user that is associated with the registered profile for the first user.

9. The computer-implemented method of claim 7, wherein the use of the one alternate name by the second user is one of:

(i) in a search query, wherein the suggestion is a result for the search query that is further based on the registered profile for the first user,

(ii) in an address field of a draft electronic message or a body of the draft electronic message, wherein the suggestion is an address for the first user from the registered profile, and

(iii) at a social network website, wherein the suggestion is a suggestion for the second user to add the first user to a group of users associated with the second user at the social network website.

10. The computer-implemented method of claim 7, further comprising:

applying, at the server, the set of patterns to the training electronic messages to extract candidate names for the training users;

selecting, at the server, a set of the candidate names having less than than a second predetermined matching accuracy with actual names in the training electronic messages to obtain a set of ambiguous names, wherein the second predetermined matching accuracy is less than the first predetermined matching accuracy; and

utilizing, at the server, the set of ambiguous names when selecting the set of alternate names for the first user by not selecting any names from the set of ambiguous names and when outputting the suggestion to the second user by not suggesting any names from the set of ambiguous names.

11. A computer-implemented method, comprising:

obtaining, at a server including one or more processors, electronic messages associated with a first user, the first user having a registered profile;

applying, at the server, a set of patterns to the electronic messages to extract candidate names for the first user, each pattern of the set of patterns including specific name context and an associated position for a name relative to the specific name context;

selecting, at the server, a set of the candidate names to obtain a set of alternate names for the first user;

storing, at the server, the set of alternate names for the first user;

detecting, at the server, a use of one alternate name from the set of alternate names by a second user at a computing device; and

outputting, from the server, a suggestion for the second user to the computing device, the suggestion being based on the registered profile for the first user.

12. The computer-implemented method of claim 11, wherein selecting the set of alternate names for the first user includes selecting candidate names having greater than a predetermined usage rate in the electronic messages to obtain the set of alternate names for the first user.

13. The computer-implemented method of claim 11, wherein the use of the one alternate name by the second user is one of:

(i) in a search query, wherein the suggestion is a result for the search query that is further based on the registered profile for the first user,

(ii) in an address field of a draft electronic message or a body of the draft electronic message, wherein the suggestion is an address for the first user from the registered profile, and

(iii) at a social network website, wherein the suggestion is a suggestion for the second user to add the first user to a group of users associated with the second user at the social network website.

14. The computer-implemented method of claim 11, further comprising:

obtaining, at the server, training electronic messages;

identifying, at the server, one or more name contexts in the training electronic messages; and

determining, at the server, candidate patterns from the name contexts, each pattern including specific name context and an associated position for a name relative to the specific name context, each candidate pattern being a candidate for the set of patterns.

15. The computer-implemented method of claim 14, further comprising:

applying, at the server, the candidate patterns to the training electronic messages to extract candidate names that correspond to the associated positions;

selecting, at the server, each candidate pattern that, when applied to the training electronic messages, extracts candidate names having greater than a first predetermined matching accuracy with actual names in the training electronic messages to obtain the set of patterns; and

storing, at the server, the set of the patterns.

16. The computer-implemented method of claim 15, wherein the training electronic messages are obtained from plurality of training users, and wherein each specific training electronic message includes at least one known name associated with a specific field of the specific training electronic message.

17. The computer-implemented method of claim 16, wherein identifying the name context in the training electronic messages includes identifying, at the server, N tokens surrounding each known name, wherein each token is a word or a punctuation, and wherein N is an integer greater than zero.

18. The computer-implemented method of claim 17, wherein determining the patterns includes determining, at the server, name context for every combination of the N tokens surrounding the known name and determining the associated position at the known name to obtain the patterns.

19. The computer-implemented method of claim 15, wherein selecting the set of patterns includes selecting each candidate pattern that, when applied to the training electronic messages, extract candidate names having greater than a first predetermined matching accuracy with actual names in the training electronic messages;

20. The computer-implemented method of claim 19, further comprising:

applying, at the server, the set of patterns to the training electronic messages to extract candidate names for the training users;

selecting, at the server, a set of the candidate names having less than a second predetermined matching accuracy with actual names in the training electronic messages to obtain a set of ambiguous names, wherein the second predetermined matching accuracy is less than the first predetermined matching accuracy; and

utilizing, at the server, the set of ambiguous names when selecting the set of alternate names for the first user by not selecting any names from the set of ambiguous names and when outputting the suggestion to the second user by not suggesting any names from the set of ambiguous names.