Communication classification and escalation using machine learning model
Systems and methods are disclosed for communication classification and escalation using a machine learning model. Metadata for a communication is obtained along with features of the content in the communication. Using a machine learning model, a likelihood that the communication comprises suspicious content or a topic of interest is determined from the metadata and or a communication transcript. If the likelihood that the communication comprises suspicious content or a topic of interest is at or above a predetermined threshold, then the communication is marked for review by a human agent; otherwise, the communication is ignored. A communication may have suspicious content or a topic of interest based upon the presence of keywords in the communication transcript either alone or in combined with metadata or other features of the content in the communication. The one or more keywords may be associated with illegal or unauthorized items, activities, or behaviors.
1. A computer-implemented method comprising:
obtaining metadata for a communication;
obtaining features of the content in the communication;
determining from the metadata and or a communication transcript, using a machine learning model, a likelihood that the communication comprises suspicious content or a topic of interest; and
providing one or more communication summaries as an input to the machine learning model, wherein the model is configured to process the communications in accordance with current values of a set of model parameters to generate as output a proposed likelihood that the communication comprises suspicious content or topic of interest.
2. The computer-implemented method of claim 1 , further comprising:
if the likelihood that the communication comprises suspicious content or topic of interest is at or above a predetermined threshold, then marking the communication for review by a human agent; and
if the likelihood that the communication comprises suspicious content or topic of interest is below a predetermined threshold, then ignoring the communication for further review.
3. The computer-implemented method of claim 1 , wherein a communication is determined to have suspicious content or topic of interest based upon the presence of one or more keywords in the communication transcript.
4. The computer-implemented method of claim 1 , wherein a communication is determined to have suspicious content or a topic of interest based upon the presence of one or more keywords in the communication transcript either alone or combined with metadata or other features of the content in the communication.
5. The computer-implemented method of claim 3 , wherein the one or more keywords are associated with illegal or unauthorized items, activities, or behaviors.
6. The computer-implemented method of claim 4 , wherein the metadata is associated with at least one of a communication parameter and a party to the communication.
7. The computer-implemented method of claim 1 , further comprising:
providing data from human agent reviews of prior communications as an input to the machine learning model, wherein the model is configured to process the communication in accordance with current values of a set of model parameters to generate as output a proposed likelihood that the communication comprises suspicious content.
8. The computer-implemented method of claim 7 , wherein the data from human agent reviews comprise one or more of new keywords assigned to prior communications, an indication whether a prior communication was escalated by the human agent, and whether the suspicious content or topic of interest produced by the machine learning system was confirmed.
9. The computer-implemented method of claim 1 , further comprising:
providing data from an end-user interface as an input to the machine learning model, wherein the model is configured to process the communications in accordance with current values of a set of model parameters to generate as output a proposed likelihood that the communication comprises suspicious content or a topic of interest.
10. The computer-implemented method of claim 9 , wherein the data from the end-user interface comprises search parameters entered by a user.
11. The computer-implemented method of claim 9 , wherein the data from the end-user interface comprises search results selected by a user.
12. The computer-implemented method of claim 1 , wherein the one or more communication summaries comprise topics or subject matter discussed in prior communications.
13. A method performed by one or more data processing apparatuses, the method comprising:
obtaining data from human agent reviews of communications recorded in one or more controlled-environment facilities; and
providing the data from human agent reviews as an input to a machine learning model, wherein the model is configured to process communications recorded in controlled-environment facilities in accordance with current values of a set of model parameters to determine a likelihood that the communications comprise suspicious content or topics of interest, wherein the communications are determined to have suspicious content or a topic of interest based upon detection of events in a participant video stream.
14. The method of claim 13 , wherein the machine learning model determines the likelihood of suspicious content based upon metadata and transcripts associated with the communications.
15. The method of claim 13 , wherein the machine learning model determines the likelihood of suspicious content based upon video images associated with the communications.
16. The method of claim 13 , further comprising:
if the likelihood that the communication comprises suspicious content or topic of interest is at or above a predetermined threshold, then sending the communication to a human agent for review; and
if the likelihood that the communication comprises suspicious content or topic of interest is below the predetermined threshold, then dropping the communication from further review.
17. The method of claim 13 , wherein the data from human agent reviews comprise one or more of new keywords assigned to prior communications, an indication whether a prior communication was escalated by the human agent, and an indication whether an expected party to a prior communication was confirmed.
18. The method of claim 13 , wherein a communication is determined to have suspicious content or a topic of interest based upon the presence of one or more keywords in a communication transcript.
19. The method of claim 13 , further comprising:
providing data from an end-user interface as an input to the machine learning model, wherein the data from the end-user interface comprises one or more of search parameters entered by a user and search results selected by the user.
20. A method performed by one or more data processing apparatuses, the method comprising:
obtaining data from human agent reviews of communications recorded in one or more controlled-environment facilities;
providing the data from human agent reviews as an input to a machine learning model, wherein the model is configured to process communications recorded in controlled-environment facilities in accordance with current values of a set of model parameters to determine a likelihood that the communications comprise suspicious content or topics of interest; and
providing one or more communication summaries as an input to the machine learning model, wherein the one or more communication summaries comprise topics or subject matter discussed in prior communications.