Use of machine-learning models in creating messages for advocacy campaigns
An advocacy system uses trained machine learning models to create messages that are sent to advocates or policymakers to achieve desired outcomes for an organization. Desired outcomes can include, for example: an advocate sending a message to a policymaker or legislative representative advocating in favor or the organization's position on an issue; a policymaker acting or voting in favor of the organization's position on an issue; or an advocate making a financial contribution to the organization. The machine learning models can be configured to select possible message characteristics or features that the system will include/use in creating/sending messages to/for individual senders and recipients. The machine learning models can be trained based on message characteristics, personal profile characteristics of senders/recipients, and outcomes from previously sent messages. Personal profile characteristics of senders/recipients can indicate correlations between certain message characteristics and certain outcomes of sending messages.
1. A method performed by a computer system having at least one processor and a memory, the method comprising, the computer system:
maintaining one or more databases that store data comprising:
a record for each person of a plurality of people;
for each person, a plurality of personal profile characteristics specific to the person;
a record for each message template of a plurality of message templates, wherein each message template comprises one or more variable fields for receiving message characteristics and one or more static fields storing content common to all messages based on the message template;
a record for each message of a plurality of messages, wherein the each message was previously sent by a sending person of the plurality of people to a receiving person of the plurality of people based on one of the plurality of message templates, the each message including a respective request by the sending person for the receiving person to perform an action; and
for each message of the plurality of messages, an indication of an outcome of the request by the sending person for the receiving person to perform the action;
training a machine learning model based on a portion of the stored data, wherein the trained machine learning model is configured to select a subset from plural message characteristics using input based on:
a message template comprising one or more variable fields for receiving message characteristics and one or more static fields storing content common to all messages based on the message template;
a set of personal profile characteristics specific to a sending person; and
a set of personal profile characteristics specific to a receiving person; and
for each pair of a plurality of pairs of a sending person and a receiving person selected from the plurality of people, the sending person being selected to send a message based on a current message template to the receiving person:
receiving a subset selected from plural message characteristics in response to providing input to the trained machine learning model, the input being based on the current message template, the set of personal profile characteristics specific to the sending person, and the set of personal profile characteristics specific to the receiving person;
integrating the selected set of message characteristics into a new message based on the current message template; and
causing the new message to be sent by or on behalf of the sending person to the receiving person.
2. The method of claim 1 , wherein the selected sending person is an advocate, wherein the selected receiving person is a policymaker, and wherein the action comprises creating, proposing, supporting or voting on legislation or policy in support of an advocacy campaign.
3. The method of claim 2 , wherein the portion of the stored data based upon which the machine learning model is trained comprises:
for each message of the plurality of messages having a record in the database:
an indication of a message template upon which the each message is based,
an indication of message characteristics included in the each message, and
the indication of an outcome of the request by the sending person for the receiving person to perform the action.
4. The method of claim 2 , further comprising:
receiving from an authoring user through a computer user interface a specification of the current message template, wherein the specification comprises an arrangement of the one or more variable fields for receiving message characteristics and the one or more static fields storing content common to all messages based on the message template, and wherein the selected message characteristics are integrated through the variable fields into the new message based on the message template.
5. The method of claim 2 , further comprising:
receiving from an authoring user through a computer user interface a specification of search criteria; and
applying the search criteria to select the plurality of pairs of a sending person and a receiving person from the plurality of people based at least on personal profile characteristics.
6. The method of claim 3 , wherein message characteristics comprise one or more of:
message topic,
requested action,
complete content of a message,
components of message content,
message formatting,
arrangement of message content within a message, and
message metadata.
7. The method of claim 3 , wherein the selected set of message characteristics indicates inclusion of certain text within a message.
8. The method of claim 7 , wherein the certain text is a facet of personal information.
9. The method of claim 8 , wherein the facet of personal information is selected from a group consisting of: biographical details, demographics, custom tags, policy proposals, and votes.
10. The method of claim 3 , wherein the selected set of message characteristics indicates exclusion of certain text within a message.
11. The method of claim 3 , wherein the selected set of message characteristics establishes a minimum duration of time since a most recent prior message was sent to the target recipient.
12. The method of claim 3 , wherein the selected set of message characteristics establishes a time of day or a day of week when the new message is to be sent.
13. The method of claim 3 , wherein personal profile characteristics comprise facets of personal information.
14. The method of claim 3 , wherein personal profile characteristics comprise one or more message characteristics associated with a requested action being performed by a recipient of a message.
15. The method of claim 2 , wherein the message characteristics, from which the trained machine learning model is configured to select a subset, consist of facets of personal information about the sending person and facets of personal information about the receiving person.
16. The method of claim 2 , wherein the message characteristics, from which the trained machine learning model is configured to select a subset, consist of facets of personal information about the sending person.
17. The method of claim 2 , wherein the message characteristics, from which the trained machine learning model is configured to select a subset, consist of facets of personal information about the receiving person.
18. The method of claim 2 , wherein the message characteristics, from which the trained machine learning model is configured to select a subset, comprise facets of personal information about the sending person and facets of personal information about the receiving person.
19. The method of claim 2 , wherein each message is sent by one or more of:
e-mail,
text message,
phone call,
social media posting,
electronic messaging through a messaging platform, and
postal mail.
20. The method of claim 1 , wherein the input used by the trained machine learning model is further based on:
a desired outcome of sending a message based on the message template to the receiving person.