Granted Patent B2

US 12,333,788 · App. 18/443,157 · Granted Jun 17, 2025

Inventors: Matthieu Portail (Mountain View, CA); Rostyslav Shevchenko (Palo Alto, CA)

Assignee: Cape Analytics, Inc.

G06V10/774G06Q50/16G06T7/0002G06T2207/20081

Abstract

Claims (39)

1. A method comprising:

determining a property;

determining property information for the property; and

determining an objective metric for a subjective characteristic of the property based on the property information using a model trained on a set of training properties associated with a set of qualitative labels for the subjective characteristic, wherein the set of training properties are ranked based on the subjective characteristic using the set of qualitative labels, and wherein the model is tuned using a qualitative label.

2. The method of claim 1 , wherein the model comprises a transformer model.

3. The method of claim 1 , wherein a qualitative label of the set of qualitative labels is manually determined by:

determining a training property pair from the set of training properties;

presenting a set of measurements of the training property pair to a user; and

wherein ranking the set of training properties comprises receiving the subjective characteristic comprising a user preference between the training property pair, wherein the user preference is determined as the qualitative label for the training property pair.

4. The method of claim 1 , further comprising determining a set of attributes for the property based on the property information, wherein the model determines the objective metric based on the set of attributes.

5. The method of claim 1 , wherein the property information comprises at least one of a set of measurements, a set of descriptions, permit data, insurance loss data, inspection data, or appraisal data.

6. The method of claim 1 , wherein the property information for the property is an input to the model, wherein the property information for the respective training property comprises a set of images and text.

7. The method of claim 1 , wherein the model is trained to predict an objective metric indicative of a ranking, determined based on the set of qualitative labels, for each of the set of training properties.

8. The method of claim 1 , wherein the model is trained by:

determining a relative ranking for each training property of the set of training properties based on the set of qualitative labels;

determining an objective metric for each training property based on the respective relative ranking; and

training the model to predict the objective metric, based on property information for the respective training property.

9. A system, comprising:

a processing system configured to:

determine a set of properties;

determine property information for each property of the set of properties; and

determine an appeal score for each property of the set of properties based on the respective property information using a model trained on a set of training properties ranked by subjective appeal using a set of qualitative labels.

10. The system of claim 9 , wherein the model comprises a transformer model.

11. The system of claim 9 , wherein a qualitative label of the set of qualitative labels is automatically determined using a comparison model.

12. The system of claim 11 , wherein the comparison model comprises a transformer model.

13. The system of claim 9 , wherein the qualitative label is further determined by:

determining a training property pair from the set of training properties;

determining measurements of the training property pair;

extracting representations for the training property pair from the measurements using an encoder; and

determining the qualitative label based on the representations using a comparison model.

14. The system of claim 9 , wherein the appeal score is an absolute score, wherein the model is trained to predict the absolute score based on relative rankings.

15. The system of claim 9 , wherein inputs to the model comprises multiple input modalities.

16. The system of claim 9 , wherein the processing system is further configured to determine explainability of the model for the appeal score.

17. The system of claim 16 , wherein determining the explainability of the model for the appeal score comprises determining a contribution of a set of attributes of the property to the appeal score.

18. The system of claim 9 , wherein the appeal score is an input to a downstream model.

19. A method comprising:

determining a property;

determining property information for the property; and

determining an objective metric for a subjective characteristic of the property based on the property information using a model trained on a set of training properties associated with a set of qualitative labels for the subjective characteristic, wherein the set of training properties are ranked based on the subjective characteristic using the set of qualitative labels.

Assignments (2)

Continuity (3)

References Cited (267)