US 12,481,669 B1
Extracting data to populate tables from natural language communications
Shinichi Kato, Kensington, CA (US); Rajesh Goli, Fremont, CA (US); Yassine Benajiba, Briarcliff Manor, NY (US); Raghavendraprasad Raghunath prasad, Redmond, WA (US); Miguel Ballesteros Martinez, New York, NY (US); Roger Scott Jenke, Seattle, WA (US); Saurabh Giri, Sammamish, WA (US); Dan Roth, Philadelphia, PA (US); Rama Krishna Sandeep Pokkunuri, Redmond, WA (US); and Atul Deo, Kirkland, WA (US)
Assigned to Amazon Technologies, Inc., Seattle, WA (US)
Filed by Amazon Technologies, Inc., Seattle, WA (US)
Filed on Jun. 30, 2022, as Appl. No. 17/810,310.
Int. Cl. G06F 15/16 (2006.01); G06F 16/22 (2019.01); G06F 16/25 (2019.01); G06F 40/40 (2020.01); G06N 20/00 (2019.01)
CPC G06F 16/254 (2019.01) [G06F 16/2282 (2019.01); G06F 40/40 (2020.01); G06N 20/00 (2019.01)] 20 Claims
OG exemplary drawing
 
1. A system, comprising:
at least one processor; and
a memory, storing program instructions that when executed by the at least one processor, cause the at least one processor to implement a data collaboration system, configured to:
capture a communication that was sent between different participants, wherein the captured communication comprises natural language data, and wherein the communication is associated with a table created at the data collaboration system;
extract one or more portions from the natural language data to update the table based on the captured communication;
determine that the captured communication includes a link to a data object that is external to the captured communication;
obtain, using the link, at least a portion of the linked data object that is external to the captured communication;
identify a content extraction component according to a data type of the linked data object that is usable to extract further content from the obtained portion of the linked data object;
generate, using a decoder of a transformer machine learning model, that decodes output as respective mapping predictions for one or more values determined based on an embedding of the table, the one or more portions extracted from the natural language data to the table, and the further content extracted from the obtained portion of the data object; and
store the one or more values into the table at locations determined based, at least in part, on the mapping predictions to update the table according to the captured communication.