US 12,249,334 B2
Systems and methods for identifying conversation roles
Steven John Graff, Grapevine, TX (US); Sayeed Khawja Mohammed, Plano, TX (US); and Devanshu D. Sheth, Dallas, TX (US)
Assigned to OPEN TEXT HOLDINGS, INC., Menlo Park, CA (US)
Filed by OPEN TEXT HOLDINGS, INC., San Mateo, CA (US)
Filed on Apr. 28, 2021, as Appl. No. 17/243,330.
Claims priority of provisional application 63/016,839, filed on Apr. 28, 2020.
Prior Publication US 2021/0335367 A1, Oct. 28, 2021
Int. Cl. G10L 15/26 (2006.01); G06F 9/54 (2006.01); G06F 40/20 (2020.01); G10L 15/00 (2013.01)
CPC G10L 15/26 (2013.01) [G06F 9/547 (2013.01); G06F 40/20 (2020.01); G10L 15/005 (2013.01)] 23 Claims
OG exemplary drawing
 
1. A method, comprising:
receiving or obtaining, by a system executing on a processor, a transcript of a conversation comprising multiple participants while the conversation is ongoing, wherein the transcript is generated from an audio of the conversation utilizing a speech-to-text recognition tool and wherein the transcript comprises text associated with the multiple participants and has no metadata describing participants of the conversation;
making, by the system, in real time, while the conversation is ongoing, an application programming interface (API) call with the transcript to a text mining engine running on an artificial intelligence platform, wherein the API call specifies a categorization functionality of the text mining engine for identifying conversation roles of the multiple participants of the conversation from the transcript comprising text associated with the multiple participants, wherein the text mining engine is a machine learning model trained using examples of conversations among people with known conversation roles, and wherein performing the categorization functionality by the text mining engine includes analyzing words in the transcript comprising text associated with the multiple participants that represent unique characteristics of the conversation and determining, based on the unique characteristics of the conversation and utilizing classification predictive modeling, a conversation role for each of the multiple participants of the conversation, wherein the text mining engine identifies individual speakers of the multiple participants by performing content analytics only on the transcript, with no further metadata describing participants of the conversation, wherein the text mining engine is configured with an automatic feedback loop to continuously provide the text mining engine with knowledge obtained from new conversations, and wherein the transcript passed to the text mining engine includes a predefined amount of initial text from each of the multiple participants;
receiving, by the system, outputs from the text mining engine in real time, while the conversation is ongoing, the outputs including identification of the conversation roles of the multiple participants of the conversation included in the transcript; and
storing, by the system in an index, the identification of the conversation roles of the participants of the conversation, wherein the index is searchable by a search engine.