Computerized system and method for multi-factor message classification and delivery
Disclosed are systems and methods for improving interactions with and between computers in content providing, searching and/or hosting systems supported by or configured with devices, servers and/or platforms. The disclosed systems and methods provide a novel messaging framework that automatically applies a multi-factor analysis technique to incoming and received messages in order to properly identify a message's type and category, which dictates the manner in which the message is displayed within a recipient's inbox. The disclosed framework operates on two levels: i) it determines whether a message is from a human or machine sender (H/M classification), and ii) it determines the messages category (MAGMA categorization).
1. A method comprising the steps of:
receiving, at a computing device on a network, a message, said message addressed to an inbox of a recipient and sent by a sender, said message comprising message content;
parsing, via the computing device, said message, and identifying message information related to the sender, the message content and a subject of the message;
applying, via the computing device, a H/M classifier to said message, said H/M classifier comprising a combination of sub-models that analyze a message based on specific features of a message, said application of the H/M classifier comprising analyzing the sender information and the content information via said combination of sub-models based on said specific features;
determining, via the computing device, based on said H/M classifier application, information indicating whether said message sender is a machine or human;
applying, via the computing device, a MAGMA classifier to said message, said MAGMA classifier comprising a word encoder for specific portions of a message, said application of the MAGMA classifier comprising analyzing the subject information and the content information via said portion specific word encoder, wherein the application of the word encoder of the MAGMA classifier comprises:
analyzing the subject information, and based on said analysis, identifying a set of words less than a predetermined threshold within said subject;
analyzing the content information, and based on said analysis, identifying a set of words less than a predetermined threshold within said content;
creating a subject vector and a content vector based on the subject and content analysis, respectively;
concatenating the subject vector and the content vector based on application of a text classifier to the subject vector and the content vector;
analyzing the concatenated vector via a text classifier, the text classifier comprising a convolutional neural network layer; and
determining the at least one specific category for the message based on said analysis;
determining, via the computing device, based on said MAGMA classifier application, information indicating at least one specific category said message is associated with;
applying, via the computing device, a set of indicators to said message, said set of indicators providing information related to said machine or human determination and said MAGMA categorization; and
communicating, via the computing device, over said network, said message to said inbox along with said set of indicators.
2. The method of claim 1 , wherein said sub-models of said H/M classifier comprise a content model, sender model, action model and salutation model.
3. The method of claim 2 , wherein application of said content model comprises:
creating a vector representation of each word in said sender information;
creating a vector representation of each word in said content information;
analyzing, via a CNN layer of the content model, each vector, and based on said analysis, concatenate each vector to a content model vector; and
analyzing said content model vector via a softmax function, and based on said analysis, determining whether said message sender was a machine or a human.
4. The method of claim 2 , wherein application of the sender model comprises:
identifying, from said sender information, an email address and a name of said sender;
generating a vector representation of said email address;
generating a vector representation of said name;
concatenating each vector based on convolutional analysis of each vector to generate a sender model vector; and
analyzing said sender model vector, and based on said analysis, determining whether said message sender was a machine or a human.
5. The method of claim 2 , wherein application of the action model comprises:
identifying data indicating actions related to message deletions within said inbox;
identifying data related to actions of opened messages within said inbox;
training a CNN model based on said identified deletion and opened data;
applying said trained CNN model to said sender information and content information; and
determining whether said message sender was a machine or a human.
6. The method of claim 2 , wherein application of the salutation model comprises:
identifying data indicating a positive label, said positive label corresponding to messages that contain an explicit salutation, said explicit salutation being a message with a match between the n beginning part of a message body and a recipient name;
identifying data indicating a negative label, said negative label corresponding to messages that do not contain said explicit salutation;
training a CNN model based on said identified positive label and negative label data;
applying said trained CNN model to at least a portion of said message content; and
determining whether said message sender was a machine or a human.
7. The method of claim 1 , wherein said at least one specific category is selected from a group of MAGMA classes comprising: Human (personal), Social, Travel, Career, Shopping, Finance and Other.
8. The method of claim 1 , wherein said communication causes said message to be displayed in a portion of a user interface of the inbox based on at least one of said set of indicators.
9. The method of claim 1 , further comprising:
searching, over a network, a datastore associated with third party entities based on information associated with said message;
identifying, based on said search, third party digital content from said datastore, said third party digital content; and
communicating, over the network, along with said communicated message for display within said inbox, said identified third party digital content.
10. A non-transitory computer-readable storage medium tangibly encoded with computer-executable instructions, that when executed by a processor associated with a computing device, performs a method comprising the steps of:
receiving, at the computing device on a network, a message, said message addressed to an inbox of a recipient and sent by a sender, said message comprising message content;
parsing, via the computing device, said message, and identifying message information related to the sender, the message content and a subject of the message;
applying, via the computing device, a H/M classifier to said message, said H/M classifier comprising a combination of sub-models that analyze a message based on specific features of a message, said application of the H/M classifier comprising analyzing the sender information and the content information via said combination of sub-models based on said specific features;
determining, via the computing device, based on said H/M classifier application, information indicating whether said message sender is a machine or human;
applying, via the computing device, a MAGMA classifier to said message, said MAGMA classifier comprising a word encoder for specific portions of a message, said application of the MAGMA classifier comprising analyzing the subject information and the content information via said portion specific word encoder, wherein the application of the word encoder of the MAGMA classifier comprises:
analyzing the subject information, and based on said analysis, identifying a set of words less than a predetermined threshold within said subject;
analyzing the content information, and based on said analysis, identifying a set of words less than a predetermined threshold within said content;
creating a subject vector and a content vector based on the subject and content analysis, respectively;
concatenating the subject vector and the content vector based on application of a text classifier to the subject vector and the content vector;
analyzing the concatenated vector via a text classifier, the text classifier comprising a convolutional neural network layer; and
determining the at least one specific category for the message based on said analysis;
determining, via the computing device, based on said MAGMA classifier application, information indicating at least one specific category said message is associated with;
applying, via the computing device, a set of indicators to said message, said set of indicators providing information related to said machine or human determination and said MAGMA categorization; and
communicating, via the computing device, over said network, said message to said inbox along with said set of indicators.
11. The non-transitory computer-readable storage medium of claim 10 , wherein said sub-models of said H/M classifier comprise a content model, sender model, action model and salutation model.
12. The non-transitory computer-readable storage medium of claim 11 , wherein application of said content model comprises:
creating a vector representation of each word in said sender information;
creating a vector representation of each word in said content information;
analyzing, via a CNN layer of the content model, each vector, and based on said analysis, concatenate each vector to a content model vector; and
analyzing said content model vector via a softmax function, and based on said analysis, determining whether said message sender was a machine or a human.
13. The non-transitory computer-readable storage medium of claim 11 , wherein application of the sender model comprises:
identifying, from said sender information, an email address and a name of said sender;
generating a vector representation of said email address;
generating a vector representation of said name;
concatenating each vector based on convolutional analysis of each vector to generate a sender model vector; and
analyzing said sender model vector, and based on said analysis, determining whether said message sender was a machine or a human.
14. The non-transitory computer-readable storage medium of claim 11 , wherein application of the action model comprises:
identifying data indicating actions related to message deletions within said inbox;
identifying data related to actions of opened messages within said inbox;
training a CNN model based on said identified deletion and opened data;
applying said trained CNN model to said sender information and content information; and
determining whether said message sender was a machine or a human.
15. The non-transitory computer-readable storage medium of claim 11 , wherein application of the salutation model comprises:
identifying data indicating a positive label, said positive label corresponding to messages that contain an explicit salutation, said explicit salutation being a message with a match between the n beginning part of a message body and a recipient name;
identifying data indicating a negative label, said negative label corresponding to messages that do not contain said explicit salutation;
training a CNN model based on said identified positive label and negative label data;
applying said trained CNN model to at least a portion of said message content; and
determining whether said message sender was a machine or a human.
16. The non-transitory computer-readable storage medium of claim 10 , wherein said at least one specific category is selected from a group of MAGMA classes comprising: Human (personal), Social, Travel, Career, Shopping, Finance and Other.
17. A computing device comprising:
a processor; and
a non-transitory computer-readable storage medium for tangibly storing thereon program logic for execution by the processor, the program logic comprising:
logic executed by the processor for receiving, at the computing device on a network, a message, said message addressed to an inbox of a recipient and sent by a sender, said message comprising message content;
logic executed by the processor for parsing, via the computing device, said message, and identifying message information related to the sender, the message content and a subject of the message;
logic executed by the processor for applying, via the computing device, a H/M classifier to said message, said H/M classifier comprising a combination of sub-models that analyze a message based on specific features of a message, said application of the H/M classifier comprising analyzing the sender information and the content information via said combination of sub-models based on said specific features;
logic executed by the processor for determining, via the computing device, based on said H/M classifier application, information indicating whether said message sender is a machine or human;
logic executed by the processor for applying, via the computing device, a MAGMA classifier to said message, said MAGMA classifier comprising a word encoder for specific portions of a message, said application of the MAGMA classifier comprising analyzing the subject information and the content information via said portion specific word encoder, wherein the application of the word encoder of the MAGMA classifier comprises:
analyzing the subject information, and based on said analysis, identifying a set of words less than a predetermined threshold within said subject;
analyzing the content information, and based on said analysis, identifying a set of words less than a predetermined threshold within said content;
creating a subject vector and a content vector based on the subject and content analysis, respectively;
concatenating the subject vector and the content vector based on application of a text classifier to the subject vector and the content vector;
analyzing the concatenated vector via a text classifier, the text classifier comprising a convolutional neural network layer; and
determining the at least one specific category for the message based on said analysis;
logic executed by the processor for determining, via the computing device, based on said MAGMA classifier application, information indicating at least one specific category said message is associated with;
logic executed by the processor for applying, via the computing device, a set of indicators to said message, said set of indicators providing information related to said machine or human determination and said MAGMA categorization; and
logic executed by the processor for communicating, via the computing device, over said network, said message to said inbox along with said set of indicators.