Systems and methods for behavior based messaging

1. A system comprising:

one or more processors; and

one or more non-transitory computer-readable storage devices storing computing instructions configured to run on the one or more processors and perform acts of:

collecting historical data of a user;

converting the historical data of the user into at least one feature vector;

calculating a first user propensity score for the user using the at least one feature vector;

calculating a second user propensity score for the user using the at least one feature vector, the second user propensity score representing a different user propensity than the first user propensity score;

normalizing the first user propensity score;

normalizing the second user propensity score;

using the first user propensity score, as normalized, to place the user into a first segment;

using the second user propensity score, as normalized, to place the user into a second segment different than the first segment; and

facilitating delivery of a message to the user based on the first segment and the second segment.

2. The system of claim 1 , wherein:

the first user propensity score is calculated with either a normal model or a strict model; and

the normal model calculates the first user propensity score for a broader first segment than the strict model.

3. The system of claim 2 , wherein the computing instructions are further configured to run on the one or more processors and perform acts of:

training the normal model on labeled training data from a plurality of metalabels; and

training the strict model on labeled training data from only one metalabel.

4. The system of claim 3 , wherein the computing instructions are further configured to run on the one or more processors and perform acts of:

facilitating a display of a graphical representation of the first segment, the second segment, the normal model, and the strict model;

receiving a selection of:

the first segment;

the second segment; and

the normal model or the strict model;

filtering a set of users into a subset of users of both the first segment and the second segment; and

facilitating delivery of messages to only the subset of users, wherein when the strict model is selected, messages delivered to the subset of users comprise information about the only one metalabel.

5. The system of claim 1 , wherein calculating the first user propensity score for the user using the at least one feature vector comprises using a logistic regression model.

6. The system of claim 5 , wherein the logistic regression model comprises maximizing an equation comprising:

)Π PP Π P P1−P, wherein:

comprises a likelihood function of or;

comprises an intercept of the logistic regression model;

comprises a coefficient vector a same size as;

comprises an index of training instances with a label of 1;

comprises an index of training instances with a label of 0;

comprises a true label of a training instance of;

comprises a predicted label of with a label of 1; and

comprises a predicted label of with a label of 0.

7. The system of claim 1 , wherein normalizing the first user propensity score comprises using a prior correction technique.

8. The system of claim 7 , wherein using the prior correction technique comprises using an equation comprising:

=

ln

⁡

[

(

1

-

τ

τ

)

⁢

(

y

¯

1

-

y

¯

)

]

,

wherein:

comprises an intercept of a logistic regression model;

τ comprises a fraction of ones in a population; and

γ comprises a fraction of ones in a sample.

9. The system of claim 1 , wherein the computing instructions are further configured to run on the one or more processors and perform acts of:

converting the at least one feature vector into a sparse representation of the at least one feature vector; and

storing the sparse representation of the at least one feature vector in a database.

10. The system of claim 1 , wherein the computing instructions are further configured to run on the one or more processors and perform acts of:

determining when the user:

interacted with the message; or

ignored the message; and

adding data to the historical data, wherein the data comprises a date and a time when the user interacted with the message or ignored the message.

11. A method being implemented via execution of computing instructions configured to run at one or more processors and configured to be stored at non-transitory computer-readable media, the method comprising:

collecting historical data of a user;

converting the historical data of the user into at least one feature vector;

calculating a first user propensity score for the user using the at least one feature vector;

calculating a second user propensity score for the user using the at least one feature vector, the second user propensity score representing a different user propensity than the first user propensity score;

normalizing the first user propensity score;

normalizing the second user propensity score;

using the first user propensity score, as normalized, to place the user into a first segment;

using the second user propensity score, as normalized, to place the user into a second segment different than the first segment; and

facilitating delivery of a message to the user based on the first segment and the second segment.

12. The method of claim 11 , wherein:

the first user propensity score is calculated with either a normal model or a strict model; and

the normal model calculates the first user propensity score for a broader first segment than the strict model.

13. The method of claim 12 , further comprising:

training the normal model on labeled training data from a plurality of metalabels; and

training the strict model on labeled training data from only one metalabel.

14. The method of claim 13 , further comprising:

facilitating a display of a graphical representation of the first segment, the second segment, the normal model, and the strict model;

receiving a selection of:

the first segment;

the second segment; and

the normal model or the strict model;

filtering a set of users into a subset of users of both the first segment and the second segment; and

facilitating delivery of messages to only the subset of users, wherein when the strict model is selected, messages delivered to the subset of users comprise information about the only one metalabel.

15. The method of claim 11 , wherein calculating the first user propensity score for the user using the at least one feature vector comprises using a logistic regression model.

16. The method of claim 15 , wherein the logistic regression model comprises maximizing an equation comprising:

)Π PP Π P P1−P, wherein:

comprises a likelihood function of or;

comprises an intercept of the logistic regression model;

comprises a coefficient vector a same size as;

comprises an index of training instances with a label of 1;

comprises an index of training instances with a label of 0;

comprises a true label of a training instance of;

comprises a predicted label of with a label of 1; and

comprises a predicted label of with a label of 0.

17. The method of claim 11 , wherein normalizing the first user propensity score comprises using a prior correction technique.

18. The method of claim 17 , wherein using the prior correction technique comprises using an equation comprising:

=

ln

⁡

[

(

1

-

τ

τ

)

⁢

(

y

¯

1

-

y

¯

)

]

,

wherein:

comprises an intercept of a logistic regression model;

τ comprises a fraction of ones in a population; and

γ comprises a fraction of ones in a sample.

19. The method of claim 11 , further comprising:

converting the at least one feature vector into a sparse representation of the at least one feature vector; and

storing the sparse representation of the at least one feature vector in a database.

20. The method of claim 11 , further comprising:

determining when the user:

interacted with the message; or

ignored the message; and

adding data to the historical data, wherein the data comprises a date and a time when the user interacted with the message or ignored the message.