US 11,989,677 B2
Framework for early warning of domain-specific events
Sanjay Rajagopalan, Palo Alto, CA (US); and Xiwei Zhou, Mountain View, CA (US)
Assigned to VIANAI SYSTEMS, INC., Palo Alto, CA (US)
Filed by ViANAI SYSTEMS, INC., Palo Alto, CA (US)
Filed on Feb. 18, 2021, as Appl. No. 17/179,197.
Prior Publication US 2022/0261732 A1, Aug. 18, 2022
Int. Cl. G06Q 10/06 (2023.01); G06F 16/35 (2019.01); G06F 40/35 (2020.01); G06N 20/00 (2019.01); G06Q 10/0635 (2023.01); G06Q 10/0637 (2023.01)
CPC G06Q 10/06375 (2013.01) [G06F 16/355 (2019.01); G06F 16/358 (2019.01); G06F 40/35 (2020.01); G06N 20/00 (2019.01); G06Q 10/0635 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method for processing unstructured data, comprising:
executing one or more trained machine learning models to match a set of candidate topics extracted from the unstructured data to a set of activities of interest for a set of domains based on (i) representations of entities, relationships, and modifiers that correspond to the set of activities of interest and are included in the one or more trained machine learning models and (ii) input that includes semantic representations of the set of candidate topics and semantic context that is included in the unstructured data and associated with the set of candidate topics;
executing one or more additional trained machine learning models to generate a set of scores for the set of activities of interest, wherein each machine learning model included in the one or more additional trained machine learning models generates one or more scores that are included in the set of scores and represent an estimated impact of one or more corresponding activities of interest on a different category within a domain;
determining one or more activities included in the set of activities of interest based on a ranking of the set of activities of interest by the set of scores;
causing one or more alerts to be outputted in a user interface, wherein each of the one or more alerts is associated with a potential event related to the one or more activities;
aggregating user feedback associated with the one or more alerts by the one or more activities and one or more categories associated with the one or more additional trained machine learning models into one or more labels, wherein each of the one or more labels represents a relevance of a corresponding activity included in the one or more activities to a category included in the one or more categories within the domain;
retraining the one or more trained machine learning models based on training data that includes the one or more labels and one or more inputs related to the one or more activities; and
for each category included in the one or more categories, retraining a corresponding machine learning model included in the one or more additional trained machine learning models based on additional training data that includes a label that is associated with the category and included in the one or more labels.