US 11,899,682 B2
Generating and presenting a searchable graph based on a graph query
Paul Pangilinan Del Villar, Bothell, WA (US); Mohamed Abdelrhman Mostafa Ali Elfeki, Renton, WA (US); Pramod Kumar Sharma, Seattle, WA (US); and Nilgoon Zarei, Kirkland, WA (US)
Assigned to Microsoft Technology Licensing, LLC, Redmond, WA (US)
Filed by Microsoft Technology Licensing, LLC, Redmond, WA (US)
Filed on Sep. 8, 2021, as Appl. No. 17/469,638.
Prior Publication US 2023/0073220 A1, Mar. 9, 2023
Int. Cl. G06F 16/26 (2019.01); G06F 16/901 (2019.01); G06F 3/04817 (2022.01)
CPC G06F 16/26 (2019.01) [G06F 3/04817 (2013.01); G06F 16/9024 (2019.01)] 20 Claims
OG exemplary drawing
 
1. A method, comprising:
identifying a collection of digital content items;
applying an entity detection model to the collection of digital content items to identify a subset of digital content items from the collection of digital content items that includes one or more entities of interest within visual content of the subset of digital content items, wherein the entity detection model is trained to detect the one or more entities of interest within a given digital content item based on a combination of metadata and visual content of the given digital content item;
providing the subset of digital content items as inputs to a concept detection model, the concept detection model being trained to predict whether one or more concepts from a collection of concepts exist within an input digital content item;
receiving a set of concept identifiers for the subset of digital content items output by the concept detection model, the set of concept identifiers indicating a plurality of concepts estimated to be present within the subset of digital content items;
determining correlation values for a plurality of correlations indicating frequencies of co-occurrences between each entity of interest from the one or more entities of interest and each concept from the plurality of concepts; and
generating a correlation graph object for the subset of digital content items, the correlation graph object including a searchable matrix including the set of concept identifiers, the one or more entities of interest, and a subset of correlations from the plurality of correlations having correlation values indicating frequency of co-occurrence between respective entities of interest and respective content identifiers within the subset of digital content items, the subset of correlations excluding a first one or more correlations having correlation scores below a minimum threshold value and a second one or more correlations having correlation scores above a maximum threshold value.