System, method, and computer program product for feature analysis using an embedding tree

Patent metadata

Field	Value
Publication number	US-12253991-B2
Application number	US-202218280828-A
Country	US
Kind code	B2
Filing date	Jun 9, 2022
Priority date	Jun 10, 2021
Publication date	Mar 18, 2025
Grant date	Mar 18, 2025

Abstract

Provided is a system for analyzing features associated with entities using an embedding tree, the system including at least one processor programmed or configured to receive a dataset associated with a plurality of entities, wherein the dataset comprises a plurality of data instances for a plurality of entities. The processor may be programmed or configured to generate at least two embeddings based on the dataset and determine split criteria for partitioning an embedding space of at least one embedding tree associated with the dataset based on feature data associated with an entity and embedding data associated with the at least two embeddings. The processor may be programmed or configured to generate at least one embedding tree having a plurality of nodes based on the split criteria. Methods and computer program products are also provided.

First claim

What is claimed is:

1. A system for analyzing features associated with entities using an embedding tree, the system comprising:
at least one processor programmed or configured to:
receive a dataset associated with a plurality of entities, wherein the dataset comprises a plurality of data instances for the plurality of entities, wherein each data instance of the plurality of data instances comprises feature data associated with an entity of the plurality of entities, and wherein the feature data comprises a plurality of feature values of a plurality of features for the entity;
generate at least two embeddings based on the dataset associated with the plurality of entities, wherein the at least two embeddings comprise embedding data associated with the at least two embeddings, and wherein the embedding data comprises values of embedding vectors of the at least two embeddings;
determine split criteria that partitions an embedding space of at least one embedding tree associated with the dataset based on the feature data associated with an entity and the embedding data associated with the at least two embeddings; and
generate the at least one embedding tree having a plurality of nodes by splitting the embedding space based on the split criteria.

2. The system of claim 1 , wherein the at least one processor is further programmed or configured to:
perform an anomaly detection process based on the embedding tree.

3. The system of claim 1 , wherein the at least one processor is further programmed or configured to:
generate an embedding for a first entity based on the at least one embedding tree.

4. The system of claim 1 , wherein, when determining the split criteria that partitions the embedding space of the at least one embedding tree associated with the dataset, the at least one processor is programmed or configured to:
determine a Bayesian Information Criterion (BIC) score for a feature associated with a Gaussian mixture model under the embedding data.

5. The system of claim 4 , wherein, when determining the BIC score for the feature associated with the Gaussian mixture model under the embedding data, the at least one processor is programmed or configured to:
for each feature of the plurality of features for the entity:
assign each feature having a value equal to 1 to a first group and assign each feature having a value equal to 0 to a second group;
determine a measure of a prior probability distribution, a measure of a mean, and a measure of a standard deviation for each of the first group and the second group using a maximum likelihood estimation (MLE) function; and
determine the BIC score based on the prior probability distribution, the measure of a mean, and the measure of a standard deviation for the first group and the prior probability distribution, the measure of a mean, and the measure of a standard deviation for the second group.

6. The system of claim 1 , wherein the at least one processor is further programmed or configured to:
display a graphical representation of the at least one embedding tree, wherein the graphical representation comprises a plurality of graphical user interface (GUI) elements associated with the plurality of nodes of the at least one embedding tree, and wherein each node of the at least one embedding tree comprises a GUI element.

7. The system of claim 5 , wherein the split criteria comprises a feature of the plurality of features, and wherein, when determining the split criteria that partitions the embedding space of the at least one embedding tree associated with the dataset, the at least one processor is programmed or configured to:
determine the feature of the plurality of features that partitions the embedding space of the at least one embedding tree associated with the dataset.

8. A method for analyzing features associated with entities using an embedding tree, the method comprising:
receiving, by at least one processor, a dataset associated with a plurality of entities, wherein the dataset comprises a plurality of data instances for the plurality of entities, wherein each data instance of the plurality of data instances comprises feature data associated with an entity of the plurality of entities, and wherein the feature data comprises a plurality of feature values of a plurality of features for the entity;
generating, by the at least one processor, at least two embeddings based on the dataset associated with the plurality of entities, wherein the at least two embeddings comprise embedding data associated with the at least two embeddings, and wherein the embedding data comprises values of embedding vectors of the at least two embeddings;
determining, by the at least one processor, split criteria that partitions an embedding space of at least one embedding tree associated with the dataset based on the feature data associated with the entity and the embedding data associated with the at least two embeddings; and
generating, by the at least one processor, the at least one embedding tree having a plurality of nodes by splitting the embedding space based on the split criteria.

9. The method of claim 8 , further comprising:
performing, by the at least one processor, an anomaly detection process based on the embedding tree.

10. The method of claim 8 , further comprising:
generating, by the at least one processor, an embedding for a first entity based on the at least one embedding tree.

11. The method of claim 8 , wherein determining the split criteria that partitions the embedding space of the at least one embedding tree associated with the dataset comprises:
determining a Bayesian Information Criterion (BIC) score for a feature associated with a Gaussian mixture model under the embedding data.

12. The method of claim 11 , wherein determining the BIC score for the feature associated with the Gaussian mixture model under the embedding data comprises:
for each feature of the plurality of features for the entity:
assigning each feature having a value equal to 1 to a first group and assigning each feature having a value equal to 0 to a second group;
determining a measure of a prior probability distribution, a measure of a mean, and a measure of a standard deviation for each of the first group and the second group using a maximum likelihood estimation (MLE) function; and
determining the BIC score based on the prior probability distribution, the measure of a mean, and the measure of a standard deviation for the first group and the prior probability distribution, the measure of a mean, and the measure of a standard deviation for the second group.

13. The method of claim 8 , further comprising:
displaying, by the at least one processor, a graphical representation of the at least one embedding tree, wherein the graphical representation comprises a plurality of graphical user interface (GUI) elements associated with the plurality of nodes of the at least one embedding tree, and wherein each node of the at least one embedding tree comprises a GUI element.

14. The method of claim 12 , wherein the split criteria comprises a feature of the plurality of features, and wherein determining the split criteria that partitions the embedding space of the at least one embedding tree associated with the dataset comprises:
determining the feature of the plurality of features that partitions the embedding space of the at least one embedding tree associated with the dataset.

15. A computer program product for analyzing features associated with entities using an embedding tree, the computer program product comprising at least one non-transitory computer-readable medium including

Assignees

• Visa Int Service Ass

Inventors

• Zheng Yan
• Zhang Wei
• Yeh Michael
• Wang Liang
• Wang Junpeng
• Jain Shubham
• Zhuang Zhongfang

Classifications

• G06F16/2246 Primary

  Trees, e.g. B+trees · CPC title

• G06F16/901 Primary

  Indexing; Data structures therefor; Storage structures (for retrieval from the web G06F16/951) · CPC title