Enhanced communication assistance with deep learning
The present disclosure provides systems and methods that leverage machine-learned models (e.g., neural networks) to provide enhanced communication assistance. In particular, the systems and methods of the present disclosure can include or otherwise leverage a machine-learned communication assistance model to detect problematic statements included in a communication and/or provide suggested replacement statements to respectively replace the problematic statements. In one particular example, the communication assistance model can include a long short-term memory recurrent neural network that detects an inappropriate tone or unintended meaning within a user-composed communication and provides one or more suggested replacement statements to replace the problematic statements.
1. A computing system to provide communication assistance, the computing system comprising:
at least one processor;
a machine-learned communication assistance model, the communication assistance model comprising:
a machine-learned context component; and
a machine-learned detection model, wherein the communication assistance model is trained to receive a first set of sequential communication data descriptive of a first communication provided by a user and, in response to receipt of the first set of sequential communication data, detect one or more problematic statements included in the first communication; and
at least one tangible, non-transitory computer-readable medium that stores instructions that, when executed by the at least one processor, cause the at least one processor to:
obtain the first set of sequential communication data descriptive of the first communication provided by the user;
input the first set of sequential communication data into the communication assistance model;
input context information about the first communication into the machine-learned context component by the at least one processor, wherein the context information input comprises data that describes one or more previous communications that led to first communication; and
receive, as an output of the communication assistance model, one or more indications that respectively identify the one or more problematic statements included in the first communication.
2. The computing system of claim 1 , wherein:
in response to receipt of the context information, the machine-learned context component generates an output that describes a context of the first set of sequential communication data; and
the machine-learned detection model generates the one or more indications based at least in part on the output of the machine-learned context component.
3. The computing system of claim 2 , wherein the context information input into the machine-learned context component comprises an identity of the user.
4. The computing system of claim 2 , wherein the context information input into the machine-learned context component comprises an identity of an intended recipient.
5. The computing system of claim 2 , wherein the output of the machine-learned context component describes an expected tone for the first communication.
6. The computing system of claim 2 , wherein the output of the machine-learned context component describes an expected language for the first communication.
7. The computing system of claim 2 , wherein the output of the machine-learned context component comprises a numerical value.
8. The computing system of claim 2 , wherein the machine-learned communication assistance model further comprises a machine-learned replacement portion that generates one or more suggested replacement statements to respectively replace the one or more problematic statements based at least in part on the output of the machine-learned context component.
9. The computing system of claim 1 , wherein the one or more previous communications that led to the first communication are contained in a same message thread as the first communication.
10. The computing system of claim 1 , wherein the communication assistance model detects offensive or derogatory language for the first communication.
11. The computing system of claim 1 , wherein the machine-learned communication assistance model comprises a sequence-to-sequence model.
12. The computing system of claim 1 , wherein:
the machine-learned communication assistance model comprises a feature extraction component;
the instructions that cause the at least one processor to input the first set of sequential communication data into the communication assistance model cause the at least one processor to input the first set of sequential communication data into the feature extraction component;
the feature extraction component extracts one or more features of the first communication; and
the communication assistance model outputs the one or more indications based at least in part on the one or more features extracted by the feature extraction component.
13. The computing system of claim 1 , wherein execution of the instructions further causes the at least one processor to select one of a plurality of different interventions based at least in part on at least one confidence score output by the machine-learned communication model, wherein each of the plurality of different interventions alerts the user to an existence of the one or more problematic statements, and wherein the plurality of different interventions have different amounts of intrusiveness.
14. The computing system of claim 1 , wherein the communication assistance model detects when an individual composing the first communication is intoxicated or otherwise incapacitated for the first communication.
15. A user computing device, the user computing device comprising:
at least one processor; and
at least one non-transitory computer-readable medium that stores instructions that, when executed by the at least one processor, cause the user computing device to:
receive a first communication provided by a user;
provide a first set of communication data descriptive of the first communication provided by the user for input into a machine-learned communication assistance model the communication assistance model comprising:
a machine-learned context component; and
a machine-learned detection model;
input context information about the first communication into the machine-learned context component by the at least one processor, wherein the context information input comprises data that describes one or more previous communications that led to first communication;
receive one or more indications that respectively identify one or more problematic statements included in the first communication, the one or more indications output by the machine-learned communication assistance model; and
display a notification that indicates the existence of the one or more problematic statements.
16. The user computing device of claim 15 , wherein:
in response to receipt of the context information, the machine-learned context component generates an output that describes a context of the first communication; and
the machine-learned detection model generates the one or more indications based at least in part on the output of the machine-learned context component.
17. The user computing device of claim 16 , wherein the context information input into the machine-learned context component comprises an identity of the user.
18. The user computing device of claim 16 , wherein the context information input into the machine-learned context component comprises an identity of an intended recipient.
19. The user computing device of claim 15 , wherein the one or more previous communications that led to the first communication are contained in a same message thread as the first communication.
20. The user computing device of claim 15 , wherein the communication assistance model detects when an individual composing the first communication is intoxicated or otherwise incapacitated for the first communication.