Dynamic conversation alerts within a communication session
Methods and systems provide for dynamic conversation alerts within a communication session. In one embodiment, the system presents, to a client device associated with a user of a communication platform, a user interface (“UI”) including a prompt for the user to submit one or more alert phrases, each alert phrase being associated with a category; receives, from the client device, a list of submitted alert phrases; and receives a transcript of a communication session between participants. For each utterance in the transcript, the system determines whether one or more predictions of relatedness are present between the utterance and one or more alert phrases from the list of submitted alert phrases. The system then transmits, to the client device, a list of related categories, each related category including one or more timestamps of utterances for which a prediction of relatedness is present for an alert phrase associated with that category.
1. A method, comprising:
presenting, to a client device associated with a user of a communication platform, a user interface (UI) comprising a prompt for the user to submit one or more alert phrases, each alert phrase being associated with a category;
receiving, from the client device, a list of submitted alert phrases;
receiving a transcript of a communication session between a plurality of participants, one of the participants being the user, the transcript comprising timestamps for a plurality of utterances associated with speaking participants;
for each utterance in the transcript, determining, at least in part via a prototypical neural network (ProtoNet), whether one or more predictions of relatedness are present between the utterance and one or more alert phrases from the list of submitted alert phrases; and
transmitting, to the client device, a list of related categories, each related category comprising one or more timestamps of utterances for which a prediction of relatedness is present for an alert phrase associated with that category.
2. The method of claim 1 , further comprising:
for each alert phrase for which a prediction of relatedness is determined to be present, determining a category associated with the alert phrase.
3. The method of claim 1 , wherein the UI further comprises a prompt for the user to submit, for each of the submitted alert phrases, a category to be associated with the alert phrase.
4. The method of claim 3 , wherein the submitted category is selected by the user from a list of prespecified categories.
5. The method of claim 3 , wherein the submitted category is created by the user.
6. The method of claim 1 , wherein the UI further comprises a prompt for the user to define at least one of the categories associated with the alert phrases.
7. The method of claim 1 , wherein determining whether the predictions of relatedness are present further comprises determining whether one or more predictions of relatedness are present between the utterance and one or more variations on alert phrases from the list of submitted alert phrases.
8. The method of claim 1 , wherein determining whether the predictions of relatedness are present is performed at least in part by one or more sentence embedding models.
9. The method of claim 1 , wherein the list of related categories with timestamps of utterances is transmitted in real-time while the user is connected to the communication session.
10. The method of claim 1 , further comprising:
generating, based on the submitted alert phrases, one or more additional alert phrases to be added to the list of submitted alert phrases, each of the additional alert phrases being associated with a category.
11. The method of claim 10 , further comprising:
segmenting each of the categories into one or more of: a positive speaker intent, a negative speaker intent, and a neutral speaker intent,
wherein each of the additional alert phrases is generated based further on the segment for a category.
12. The method of claim 1 , further comprising:
detecting that one of the alert phrases has been associated with a category that differs in intent from the alert phrase; and
associating the alert phrase with a different category.
13. The method of claim 1 , further comprising:
detecting that one of the submitted alert phrases at least partially matches an existing alert phrase; and
executing a target action in response to the detection, the target action comprising one or more of: removing the alert phrase from the list of submitted phrases, prompting the user to submit a different alert phrase, and replacing the submitted alert phrase with a generated alert phrase.
14. A communication system comprising one or more processors configured to perform operations of:
presenting, to a client device associated with a user of a communication platform, a user interface (UI) comprising a prompt for the user to submit one or more alert phrases, each alert phrase being associated with a category;
receiving, from the client device, a list of submitted alert phrases;
receiving a transcript of a communication session between a plurality of participants, one of the participants being the user, the transcript comprising timestamps for a plurality of utterances associated with speaking participants;
for each utterance in the transcript, determining, at least in part via a prototypical neural network (ProtoNet), whether one or more predictions of relatedness are present between the utterance and one or more alert phrases from the list of submitted alert phrases; and
transmitting, to the client device, a list of related categories, each related category comprising one or more timestamps of utterances for which a prediction of relatedness is present for an alert phrase associated with that category.
15. The communication system of claim 14 , wherein determining whether the predictions of relatedness are present is performed at least in part by one or more intent detection algorithms.
16. The communication system of claim 14 , wherein determining whether the predictions of relatedness are present is performed at least in part using one or more of: few-shot detection techniques, and zero-shot detection techniques.
17. The communication system of claim 14 , wherein the ProtoNet comprises presenting a new task with unseen inputs and unseen classes.
18. The communication system of claim 14 , wherein the list of related categories with timestamps of utterances is transmitted in real-time while the user is connected to the communication session.
19. The communication system of claim 14 , wherein determining whether the predictions of relatedness are present is performed at least in part by one or more pre-trained language learning models.
20. A non-transitory computer-readable medium comprising instructions, that when executed by one or more processors, causes the one or more processors to perform operations comprising:
presenting, to a client device associated with a user of a communication platform, a user interface (UI) comprising a prompt for the user to submit one or more alert phrases, each alert phrase being associated with a category;
receiving, from the client device, a list of submitted alert phrases;
receiving a transcript of a communication session between a plurality of participants, one of the participants being the user, the transcript comprising timestamps for a plurality of utterances associated with speaking participants;
for each utterance in the transcript, determining, at least in part via a prototypical neural network (ProtoNet), whether one or more predictions of relatedness are present between the utterance and one or more alert phrases from the list of submitted alert phrases; and
transmitting, to the client device, a list of related categories, each related category comprising one or more timestamps of utterances for which a prediction of relatedness is present for an alert phrase associated with that category.