| CPC G06V 20/41 (2022.01) [G06F 18/22 (2023.01); G06F 40/30 (2020.01); G06V 10/426 (2022.01); G06V 10/44 (2022.01); G06V 10/70 (2022.01); G06V 10/761 (2022.01); G06V 10/774 (2022.01); G06V 10/7753 (2022.01); G06V 10/86 (2022.01); G06V 20/46 (2022.01); G06V 20/70 (2022.01); G06V 30/274 (2022.01); G06V 40/20 (2022.01); G10L 25/57 (2013.01)] | 18 Claims |
|
1. A method comprising:
acquiring training video data that portrays a plurality of training interacting events;
labeling each training interacting event of the plurality of training interacting events as positive or negative to create a plurality of positive interacting events and a plurality of negative interacting events;
creating a plurality of positive training relationship graphs by, for each positive interacting event of the plurality of positive interacting events:
extracting positive training image data, positive training audio data, and positive training semantic text data from the training video data;
analyzing, by a first computer-implemented machine learning model, at least one of the positive training image data, the positive training audio data, and the positive training semantic text data to identify a plurality of positive training video features; and
analyzing the plurality of positive training video features to create a positive relationship graph, wherein the positive relationship graph includes a plurality of positive training nodes and a plurality of positive training edges extending between nodes of the plurality of positive training nodes;
creating a plurality of negative training relationship graphs by, for each negative interacting event of the plurality of negative interacting events:
extracting negative training image data, negative training audio data, and negative semantic text data from the training video data;
analyzing, by the first computer-implemented machine learning model, at least one of the negative training image data, the negative training audio data, and the negative semantic text data to identify a plurality of negative training video features;
analyzing the plurality of negative training video features to create a negative relationship graph, wherein the negative relationship graph includes a plurality of negative training nodes and a plurality of negative training edges extending between nodes of the plurality of negative training nodes;
analyzing the plurality of positive training relationship graphs to identify a plurality of positive graph features;
analyzing the plurality of negative training relationship graphs to identify a plurality of negative graph features;
training a second computer-implemented machine learning model to identify positive and negative interacting events using the plurality of positive graph features and the plurality of negative graph features;
identifying a first key feature using the trained second computer-implemented machine learning model;
acquiring digital video data that portrays an interacting event, the interacting event comprising a plurality of interactions between a plurality of individuals;
extracting image data, audio data, and semantic text data from the video data;
analyzing, by the first computer-implemented machine learning model, at least one of the image data, the audio data, and the semantic text data to identify a plurality of video features;
analyzing the plurality of video features to create a relationship graph, wherein:
the relationship graph comprises a plurality of nodes and a plurality of edges;
each node of the plurality of nodes represents an individual of the plurality of individuals;
each edge of the plurality of edges extends between two nodes of the plurality of nodes; and
the plurality of edges represents the plurality of interactions;
determining whether the first key feature is present in the relationship graph, wherein presence of the first key feature is predictive of a positive outcome of the interacting event; and
outputting, by a user interface, an indication whether the first key feature is present in the relationship graph.
|