Classifying electronic messages using individualized artificial intelligence techniques

• US 9,946,789 B1
• Filed: 04/28/2017
• Issued: 04/17/2018
• Est. Priority Date: 04/28/2017
• Status: Expired due to Fees

• Current Assignee: Shenzhen Cestbon Technology Co. Limited
• Original Assignee: Shenzhen Cestbon Technology Co. Limited
• Inventors: Li, Kedan
• Primary Examiner(s): Ly, Cheyne D
• Application Number:
  US15/582,571
• Time in Patent Office: 354 Days
• Field of Search: 707737, 715702, 715752
• CPC Class Codes: G06F 16/35 Clustering; Classification
• CPC Class Codes: G06F 16/93 Document management systems
• CPC Class Codes: G06F 40/284 Lexical analysis, e.g. tokenisation or collocates
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06Q 10/107 Computer-aided management of electronic mailing [e-mailing]
• CPC Class Codes: H04L 51/212 using filtering or selective blocking
• CPC Class Codes: H04L 51/42 Mailbox-related aspects, e.g. synchronisation of mailboxes

Abstract

An example method includes: identifying message labels for electronic messages; identifying, based on a classification model specific to a first user, a first and a second message labels for a first and a second electronic messages; detecting a user action by the first user on the second electronic message to indicate the first message label is descriptive of the second electronic message; responsive to the user action, re-training the classification model based on tokens produced from the second electronic message to produce an updated classification model specific to the first user; after re-training is completed, detecting an incoming electronic message having a timestamp later in time than timestamps for the first and the second electronic messages; determining that the incoming electronic message shares a predefined number of tokens with the second electronic message; and assigning, based on the updated classification model, the first message label to the incoming electronic message.

Claims

1. A system, comprising:

a non-transitory memory; and

one or more hardware processors coupled to the non-transitory memory and configured to execute instructions to perform operations comprising;

obtaining a plurality of electronic messages associated with a user, the plurality of electronic messages including a first electronic message and a second electronic message;

identifying a plurality of message labels associated with the plurality of electronic messages, the plurality of message labels including a first message label and a second message label;

identifying, based on a classification model specific to a first user, a first message label associated with the first electronic message and a second message label associated with the second electronic message;

producing one or more tokens from the second electronic message;

detecting a finger gesture by the first user on the second electronic message to apply the first message label to the second electronic message;

responsive to detecting the finger gesture, re-training the classification model using a computer based on the one or more tokens produced from the second electronic message to produce an updated classification model specific to the first user;

re-training the classification model comprising;

applying one or more natural language processing techniques to the plurality of electronic messages to produce a plurality of tokens;

generating a first message cluster and a second message cluster based on the plurality of tokens, the first message cluster including one or more electronic messages sharing a predefined number of similarities with tokens produced from the second electronic message;

assigning, the first message label, to messages included in the first message cluster; and

updating the classification model based on the first message label and the messages included in the first message cluster;

after re-training the classification model is completed, detecting an incoming electronic message not included in the plurality of electronic messages, the incoming electronic message being associated with a timestamp that is later in time than timestamps associated with the first electronic message and the second electronic message;

comparing the incoming electronic message with the one or more tokens produced from the second electronic message;

determining that the incoming electronic message shares a predefined number of similarities with the one or more tokens produced from the second electronic message; and

applying, based on the updated classification model, the first message label to the incoming electronic message.

2. The system of claim 1, wherein the first message label is applied to the incoming electronic message based exclusively on the updated classification model.

3. The system of claim 1, wherein the operations further comprise:

determining that one or more second classification models are specific to a second user different from the first user; and

re-training the one or more second classification models without regard to the finger gesture by the first user on the second electronic message.

4. The system of claim 1, wherein the first message cluster and the second message cluster are generated based on a density based clustering method or a distribution based clustering method.

5. The system of claim 1, wherein the operations further comprise:

maintaining a single message label for the first message cluster.

6. The system of claim 1, wherein the operations further comprise:

splitting an existing message cluster into two or more message clusters, the two or more message clusters including the first message cluster and the second message cluster.

7. The system of claim 1, wherein applying the first message label to messages included in the first message cluster is based on a distribution of message labels associated with messages in the first message cluster and a time-biased function that gives more weight to an electronic messaging having a more recent timestamp.

8. The system of claim 1, wherein the operations further comprise:

extracting text-based content from the plurality of electronic messages; and

applying the one or more natural language processing techniques to the extracted text-based content.

9. The system of claim 8, wherein the plurality of electronic messages include video content or audio content.

10. A method comprising:

obtaining a plurality of electronic messages associated with a user, the plurality of electronic messages including a first electronic message and a second electronic message;

identifying a plurality of message labels associated with the plurality of electronic messages, the plurality of message labels including a first message label and a second message label;

identifying, based on a classification model specific to a first user, a first message label associated with the first electronic message and a second message label associated with the second electronic message;

producing one or more tokens from the second electronic message;

detecting a finger gesture by the first user on the second electronic message to apply the first message label to the second electronic message;

responsive to detecting the finger gesture, re-training the classification model using a computer based on the one or more tokens produced from the second electronic message to produce an updated classification model specific to the first user;

re-training the classification model comprising;

applying one or more natural language processing techniques to the plurality of electronic messages to produce a plurality of tokens;

generating a first message cluster and a second message cluster based on the plurality of tokens, the first message cluster including one or more electronic messages sharing a predefined number of similarities with tokens produced from the second electronic message;

assigning, the first message label, to messages included in the first message cluster; and

updating the classification model based on the first message label and the messages included in the first message cluster;

after re-training the classification model is completed, detecting an incoming electronic message not included in the plurality of electronic messages, the incoming electronic message being associated with a timestamp that is later in time than timestamps associated with the first electronic message and the second electronic message;

comparing the incoming electronic message with the one or more tokens produced from the second electronic message;

determining that the incoming electronic message shares a predefined number of similarities with the one or more tokens produced from the second electronic message; and

applying, based on the updated classification model, the first message label to the incoming electronic message.

11. The method of claim 10, wherein the first message label is applied to the incoming electronic message based exclusively on the updated classification model.

12. The method of claim 10, further comprising:

determining that one or more second classification models are specific to a second user different from the first user; and

re-training the one or more second classification models without regard to the finger gesture by the first user on the second electronic message.

13. The method of claim 10, wherein the first message cluster and the second message cluster are generated based on a density based clustering method or a distribution based clustering method.

14. The method of claim 10, further comprising:

maintaining a single message label for the first message cluster.

15. The method of claim 10, further comprising:

splitting an existing message cluster into two or more message clusters, the two or more message clusters including the first message cluster and the second message cluster.

16. The method of claim 10, wherein applying the first message label to messages included in the first message cluster is based on a distribution of message labels associated with messages in the first message cluster and a time-biased function that gives more weight to an electronic messaging having a more recent timestamp.

17. A non-transitory computer readable storage medium storing one or more programs, the one or more programs comprising instructions, which when executed by a computing system with one or more processors, cause the computing system to execute a method of:

obtaining a plurality of electronic messages associated with a user, the plurality of electronic messages including a first electronic message and a second electronic message;

identifying a plurality of message labels associated with the plurality of electronic messages, the plurality of message labels including a first message label and a second message label;

identifying, based on a classification model specific to a first user, a first message label associated with the first electronic message and a second message label associated with the second electronic message;

producing one or more tokens from the second electronic message;

detecting a finger gesture by the first user on the second electronic message to apply the first message label to the second electronic message;

responsive to detecting the finger gesture, re-training the classification model using a computer based on the one or more tokens produced from the second electronic message to produce an updated classification model specific to the first user;

re-training the classification model comprising;

applying one or more natural language processing techniques to the plurality of electronic messages to produce a plurality of tokens;

generating a first message cluster and a second message cluster based on the plurality of tokens, the first message cluster including one or more electronic messages sharing a predefined number of similarities with tokens produced from the second electronic message;

assigning, the first message label, to messages included in the first message cluster; and

updating the classification model based on the first message label and the messages included in the first message cluster;

after re-training the classification model is completed, detecting an incoming electronic message not included in the plurality of electronic messages, the incoming electronic message being associated with a timestamp that is later in time than timestamps associated with the first electronic message and the second electronic message;

comparing the incoming electronic message with the one or more tokens produced from the second electronic message;

determining that the incoming electronic message shares a predefined number of similarities with the one or more tokens produced from the second electronic message; and

applying, based on the updated classification model, the first message label to the incoming electronic message.

18. The non-transitory computer readable storage medium of claim 17, the method further comprising:

determining that one or more second classification models are specific to a second user different from the first user; and

re-training the one or more second classification models without regard to the finger gesture by the first user on the second electronic message.

19. The non-transitory computer readable storage medium of claim 17, wherein the first message cluster and the second message cluster are generated based on a density based clustering method or a distribution based clustering method.

20. The non-transitory computer readable storage medium of claim 17, wherein applying the first message label to messages included in the first message cluster is based on a distribution of message labels associated with messages in the first message cluster and a time-biased function that gives more weight to an electronic messaging having a more recent timestamp.