System and method for message reaction analysis

1. A method comprising:

obtaining reaction data for each of a plurality of communications, the reaction data for each communication characterizing a reaction of a corresponding first user to a communication sent by a corresponding second user using a corresponding communication service from a plurality of communication services, wherein each of the plurality of communication services allows users to react to received communications by selecting from a predetermined set of proprietary reactions that are supported by the communication service, wherein different communication services in the plurality of communication services have different sets of proprietary reactions that are supported by the communication service, and wherein the corresponding first user has not selected any proprietary reaction supported by the corresponding communication service in response to the communication sent by the corresponding second user;

for each of the plurality of communications:

analyzing the reaction data for the communication to generate standardized reaction data that characterizes a sentiment of the reaction of the corresponding first user to the communication, wherein the standardized reaction data comprises multiple reaction types and a respective associated confidence score for each reaction type, wherein the same set of multiple reaction types are used for all of the plurality of communication services;

mapping, using a machine learning model, the standardized reaction data that characterizes the sentiment of the reaction of the corresponding first user to one or more proprietary reactions from the predetermined set of proprietary reactions that are supported by the corresponding communication service, wherein the machine learning model has been trained on a training data set from communication history data of the corresponding communication service to receive as input the standardized reaction data and to generate as output a score distribution over propriety reactions from the predetermined set of proprietary reactions that are supported by the corresponding communication service; and

providing, to the corresponding communication service, data identifying the one or more proprietary reactions.

2. The method of claim 1 , wherein the communication sent by the corresponding second user comprises one or more text messages.

3. The method of claim 1 , wherein the reaction data for the communication sent by the corresponding second user comprises reaction data captured using a plurality of different modalities, and wherein generating the standardized reaction data comprises:

analyzing, for each of the different modalities, the reaction data captured by the corresponding modality to generate modality-specific reaction data, wherein the modality-specific reaction data comprises one or more reaction data types and associated confidence scores; and

combining the modality-specific reaction data for the plurality of modalities to generate the standardized reaction data.

4. The method of claim 3 , wherein the reaction data types include biometric data, video data, or audio data.

5. The method of claim 1 , wherein the standardized reaction data comprises a respective confidence score for each of a plurality of global reaction types, and wherein mapping the standardized reaction data to the one or more proprietary reactions that are supported by the corresponding communication service comprises:

obtaining a set of rules that define mappings from global reaction types and confidence scores to proprietary reactions; and

mapping the standardized reaction data to the one or more proprietary reactions that are supported by the corresponding communication service based on the set of rules.

6. The method of claim 1 , wherein the training data set from the communication history data of the corresponding communication service comprises past communications where one or more users explicitly selected proprietary reactions.

7. The method of claim 1 , further comprising:

providing, to the communication service, the one or more proprietary reactions that are supported by the communication service for injecting into a communications session between the corresponding first user and the corresponding second user using the communication service; and

indexing the communication sent by the corresponding second user using the data identifying the one or more proprietary reactions.

8. The method of claim 1 wherein the training data set includes:

(i) standardized reaction data for a plurality of past communications of each of the plurality of communication service; and

(ii) for each past communication of the plurality of past communications, a proprietary reaction that was explicitly selected in response to the past communication.

9. A system comprising one or more computers and one or more storage devices storing instructions that when executed by the one or more computers cause the one or more computers to perform operations comprising:

obtaining reaction data for each of a plurality of communications, the reaction data for each communication characterizing a reaction of a corresponding first user to a communication sent by a corresponding second user using a corresponding communication service from a plurality of communication services, wherein each of the plurality of communication services allows users to react to received communications by selecting from a predetermined set of proprietary reactions that are supported by the communication service, wherein different communication services in the plurality of communication services have different sets of proprietary reactions that are supported by the communication service, and wherein the corresponding first user has not selected any proprietary reaction supported by the corresponding communication service in response to the communication sent by the corresponding second user;

for each of the plurality of communications:

analyzing the reaction data for the communication to generate standardized reaction data that characterizes a sentiment of the reaction of the corresponding first user to the communication, wherein the standardized reaction data comprises multiple reaction types and a respective associated confidence score for each reaction type, wherein the same set of multiple reaction types are used for all of the plurality of communication services;

mapping, using a machine learning model, the standardized reaction data that characterizes the sentiment of the reaction of the corresponding first user to one or more proprietary reactions from the predetermined set of proprietary reactions that are supported by the corresponding communication service, wherein the machine learning model has been trained on a training data set from communication history data of the corresponding communication service to receive as input the standardized reaction data and to generate as output a score distribution over propriety reactions from the predetermined set of proprietary reactions that are supported by the corresponding communication service; and

providing, to the corresponding communication service, data identifying the one or more proprietary reactions.

10. The system of claim 9 , wherein the communication sent by the corresponding second user comprises one or more text messages.

11. The system of claim 9 , wherein the reaction data for the communication sent by the corresponding second user comprises reaction data captured using a plurality of different modalities, and wherein generating the standardized reaction data comprises:

analyzing, for each of the different modalities, the reaction data captured by the corresponding modality to generate modality-specific reaction data, wherein the modality-specific reaction data comprises one or more reaction data types and associated confidence scores; and

combining the modality-specific reaction data for the plurality of modalities to generate the standardized reaction data.

12. The system of claim 9 , wherein the standardized reaction data comprises a respective confidence score for each of a plurality of global reaction types, and wherein mapping the standardized reaction data to the one or more proprietary reactions that are supported by the corresponding communication service comprises:

obtaining a set of rules that define mappings from global reaction types and confidence scores to proprietary reactions; and

mapping the standardized reaction data to the one or more proprietary reactions that are supported by the corresponding communication service based on the set of rules.

13. The system of claim 9 , wherein the training data set from the communication history data of the corresponding communication service comprises past communications where one or more users explicitly selected proprietary reactions.

14. The system of claim 9 , the operations further comprising:

providing, to the communication service, the one or more proprietary reactions that are supported by the communication service for injecting into a communications session between the corresponding first user and the corresponding second user using the communication service; and

indexing the communication sent by the corresponding second user using the data identifying the one or more proprietary reactions.

15. One or more non-transitory computer storage medium encoded with a computer program, the computer program storing instructions that when executed by one or more computers cause the one or more computers to perform operations comprising:

obtaining reaction data for each of a plurality of communications, the reaction data for each communication characterizing a reaction of a corresponding first user to a communication sent by a corresponding second user using a corresponding communication service from a plurality of communication services, wherein each of the plurality of communication services allows users to react to received communications by selecting from a predetermined set of proprietary reactions that are supported by the communication service, wherein different communication services in the plurality of communication services have different sets of proprietary reactions that are supported by the communication service, and wherein the corresponding first user has not selected any proprietary reaction supported by the corresponding communication service in response to the communication sent by the corresponding second user;

for each of the plurality of communications:

analyzing the reaction data for the communication to generate standardized reaction data that characterizes a sentiment of the reaction of the corresponding first user to the communication, wherein the standardized reaction data comprises multiple reaction types and a respective associated confidence score for each reaction type, wherein the same set of multiple reaction types are used for all of the plurality of communication services;

mapping, using a machine learning model, the standardized reaction data that characterizes the sentiment of the reaction of the corresponding first user to one or more proprietary reactions from the predetermined set of proprietary reactions that are supported by the corresponding communication service, wherein the machine learning model has been trained on a training data set from communication history data of the corresponding communication service to receive as input the standardized reaction data and to generate as output a score distribution over propriety reactions from the predetermined set of proprietary reactions that are supported by the corresponding communication service; and

providing, to the corresponding communication service, data identifying the one or more proprietary reactions.

16. The non-transitory computer storage medium of claim 15 , wherein the communication sent by the corresponding second user comprises one or more text messages.

17. The non-transitory computer storage medium of claim 15 , wherein the reaction data for the communication sent by the corresponding second user comprises reaction data captured using a plurality of different modalities, and wherein generating the standardized reaction data comprises:

analyzing, for each of the different modalities, the reaction data captured by the corresponding modality to generate modality-specific reaction data, wherein the modality-specific reaction data comprises one or more reaction data types and associated confidence scores; and

combining the modality-specific reaction data for the plurality of modalities to generate the standardized reaction data.

18. The non-transitory computer storage medium of claim 15 , wherein the standardized reaction data comprises a respective confidence score for each of a plurality of global reaction types, and wherein mapping the standardized reaction data to the one or more proprietary reactions that are supported by the corresponding communication service comprises:

obtaining a set of rules that define mappings from global reaction types and confidence scores to proprietary reactions; and

mapping the standardized reaction data to the one or more proprietary reactions that are supported by the corresponding communication service based on the set of rules.

19. The non-transitory computer storage medium of claim 15 , wherein the training data set from the communication history data of the corresponding communication service comprises past communications where one or more users explicitly selected proprietary reactions.

20. The method of claim 15 , further comprising:

providing, to the communication service, the one or more proprietary reactions that are supported by the communication service for injecting into a communications session between the corresponding first user and the corresponding second user using the communication service; and

indexing the communication sent by the corresponding second user using the data identifying the one or more proprietary reactions.