Granted Patent B2

US 12,579,786 · App. 18/370,758 · Granted Mar 17, 2026

Inventors: Sarah Cebulski (Palo Alto, CA); Kyler J. Brown (Palo Alto, CA)

Assignee: Cape Analytics, Inc.

G06V10/7715G06V10/761G06V10/764G06V10/82G06F18/24G06N3/08G06Q50/16G06T7/00G06V10/25

Abstract

Claims (29)

1 . A method, comprising:

determining a set of images associated with a property;

determining a text representation for the property using a trained machine learning model, comprising:

transforming the set of images into a learned embedding space; and

determining the text representation for the property based on the transformed set of images,

wherein the trained machine learning model comprises an encoder trained using self-supervised learning;

for each of a set of reference properties, determining a text representation for the reference property; and

determining a typicality metric for the property based on the text representation for the property and the text representations for the reference properties.

2 . The method of claim 1 , wherein the typicality metric is determined using a second trained machine learning model.

3 . The method of claim 2 , wherein the second trained machine learning model is trained using automated valuation model data for each of a set of the training properties.

4 . The method of claim 1 , wherein transforming the set of images into the learned embedding space comprises extracting a set of attribute values from the set of images and transforming the set of attribute values into the learned embedding space, wherein the text representation for the property is determined based on the transformed set of attribute values.

5 . The method of claim 1 , further comprising determining an explanation associated with the typicality metric based on the text representation for the property and the text representations for the reference properties.

6 . The method of claim 1 , wherein the text representation for the property is further determined based on a parcel boundary for the property.

7 . The method of claim 1 , wherein the typicality metric is determined based on a comparison between the text representation for the property and the text representations for the reference properties.

8 . The method of claim 7 , further comprising aggregating the text representations for the reference properties, wherein the comparison comprises a comparison between the text representation for the property and the aggregated text representations.

9 . A system, comprising:

a processing system configured to:

using a learned embedding space, determine a property representation based on property information for a property, wherein the property information comprises images;

determining a reference representation using the learned embedding space, wherein the reference representation is associated with a set of reference properties; and

using a trained machine learning model, determine a typicality metric for the property based on the property representation and the reference representation,

wherein the trained machine learning model comprises an encoder trained using self-supervised learning.

10 . The system of claim 9 , wherein the property representation comprises text.

11 . The system of claim 9 , wherein the processing system is further configured to determine an explanation associated with the typicality metric based on the property representation and the reference representation.

12 . The system of claim 9 , wherein the processing system is further configured to extract a set of attribute values from the property information, wherein determining the property representation comprises transforming the set of attribute values into the learned embedding space using a second trained machine learning model, wherein the property representation is determined based on the transformed set of attribute values.

13 . The system of claim 9 , wherein the trained machine learning model is trained using automated valuation model data for each of a set of the training properties.

14 . The system of claim 13 , wherein the automated valuation model data comprises automated valuation model error.

15 . The system of claim 9 , wherein the property information further comprises at least one of: a property description, insurance data, appraisal data, or valuation information.

16 . The system of claim 9 , wherein the property information is retrieved from database.

17 . The system of claim 9 , wherein a set of comparable properties is selected from a set of properties using the typicality metric.

Assignments (1)

Continuity (3)

References Cited (264)