Granted Patent B2

US 11,238,473 · App. 16/533,489 · Granted Feb 1, 2022

Inventors: Stephen Milton (Lyons, CO); Duncan McCall (Greenwhich, CT)

Assignee: PlaceIQ, Inc.

G06Q30/0205G06Q30/0201G06Q30/0204H04W4/029G06Q30/02

Abstract

Claims (127)

1. A method of discovering psychographic segments of consumers with unsupervised machine learning, the method comprising:

obtaining, with one or more computers, a first set of more than 10,000 consumer-behavior records, each consumer-behavior record indicating, for a respective consumer, at least two businesses patronized by the respective consumer;

converting, with one or more computers, the first set of consumer-behavior records into respective consumer-behavior vectors in a first set of consumer-behavior vectors;

determining, with one or more computers, more than 5 psychographic segments of consumers by training an unsupervised machine learning model based on the first set of consumer-behavior vectors, wherein the first set of consumer-behavior vectors are not pre-labeled as members of any of the psychographic segments before determining the psychographic segments;

obtaining, with one or more computers, a second set of consumer-behavior records after determining the psychographic segments of consumers;

converting, with one or more computers, the second set of consumer-behavior records into respective consumer-behavior vectors in a second set of consumer-behavior vectors;

classifying, with one or more computers, the second set of consumer-behavior vectors as each belonging to at least a respective one of psychographic segments with the trained machine learning model;

determining, with one or more computers, based on the classification, for at least some of the second set of consumer behavior records, a property associated with a corresponding one of the psychographic segments and not associated with the respective consumer; and

predicting, with one or more computers, based on the classification, for each of the second set of consumer behavior records, a likelihood of the respective consumer engaging in behavior associated with a corresponding one of the psychographic segments, wherein predicting a likelihood of the respective consumer engaging in behavior associated with a corresponding one of the psychographic segments comprises predicting a likelihood to patronize a retail store patronized by others in the same segment.

2. The method of claim 1 , wherein determining more than 5 psychographic segments of consumers comprises:

clustering the first set of consumer-behavior vectors based on pairwise distance in a parameter space defined by attributes of the consumer-behavior records.

3. The method of claim 2 , wherein clustering the first set of consumer behavior vectors comprises determining that at least some vectors have more than a threshold number of other vectors within a threshold distance in the parameter space.

4. The method of claim 3 , wherein determining more than 5 psychographic segments of consumers comprises:

selecting N cluster centers for N candidate clusters from among the first set of vectors, where N is an integer greater than or equal to 5;

assigning vectors in the first set to the candidate clusters based on distance from the respective cluster center; and

iteratively updating the cluster centers to reduce an aggregate measure of distances from the vectors to cluster centers and re-assigning the vectors to updated clusters until less than a threshold amount of vectors change clusters between iterations.

5. The method of claim 1 , wherein determining more than 5 psychographic segments of consumers comprises:

determining at least one segment with at least one latent variable model based on a mixture model of the first set of consumer-behavior vectors.

6. The method of claim 1 , wherein determining more than 5 psychographic segments of consumers comprises:

forming a 2 or higher dimensional lattice of nodes, each node having a node vector and a coordinate in the lattice, the node vector having the same number of dimensions as the vectors in the first set of consumer-behavior records;

initializing the node vectors;

selecting a given consumer-behavior vector from among the first set of consumer-behavior vectors;

for each of at least half the nodes, determining a matching node having an associated node vector that is most similar to the given consumer-behavior vector;

determining adjacent nodes within a threshold distance in the lattice; and

adjusting node vectors of the adjacent nodes to be more similar to the given consumer-behavior vector.

7. The method of claim 1 , wherein:

determining more than 5 psychographic segments of consumers by training an unsupervised machine learning model based on the consumer-behavior vectors comprises:

determining at least 5 corresponding envelopes in a parameter space of the first set of consumer behavior records;

classifying the second set of consumer-behavior vectors as each belonging to at least a respective one of psychographic segments with the trained machine learning model comprises:

determining which parameter space envelope contains the respective member of the second set of consumer-behavior vectors.

8. The method of claim 1 , wherein predicting a likelihood of the respective consumer engaging in behavior associated with a corresponding one of the psychographic segments comprises:

predicting a likelihood to transition from patronizing one plurality of retail businesses to patronizing another plurality of retail businesses transitioned to by others in the same segment.

9. The method of claim 1 , wherein predicting a likelihood of the respective consumer engaging in behavior associated with a corresponding one of the psychographic segments comprises:

predicting a likelihood to consume a product consumed by others in the same segment.

10. The method of claim 1 , wherein determining more than 5 psychographic segments of consumers by training an unsupervised machine learning model with the consumer-behavior vectors comprises performing steps for training an unsupervised machine learning model.

11. The method of claim 1 , comprising:

selecting content based on the property associated with the corresponding one of the psychographic segments and not associated with the respective consumer; and

causing the content to be sent to a user.

12. A non-transitory, machine-readable media storing instructions that when executed by one or more computers effectuate operations comprising:

obtaining, with one or more computers, a first set of more than 10,000 consumer-behavior records, each consumer-behavior record indicating, for a respective consumer, at least two businesses patronized by the respective consumer;

converting, with one or more computers, the first set of consumer-behavior records into respective consumer-behavior vectors in a first set of consumer-behavior vectors;

determining, with one or more computers, more than 5 psychographic segments of consumers by training an unsupervised machine learning model with the first set of consumer-behavior vectors, wherein the first set of consumer-behavior vectors are not pre-labeled as members of any of the psychographic segments before determining the psychographic segments;

obtaining, with one or more computers, a second set of consumer-behavior records after determining the psychographic segments of consumers;

converting, with one or more computers, the second set of consumer-behavior records into respective consumer-behavior vectors in a second set of consumer-behavior vectors;

classifying, with one or more computers, the second set of consumer-behavior vectors as each belonging to at least a respective one of psychographic segments with the trained machine learning model;

determining, with one or more computers, based on the classification, for at least some of the second set of consumer behavior records, a property associated with a corresponding one of the psychographic segments and not associated with the respective consumer; and

predicting, with one or more computers, based on the classification, for each of the second set of consumer behavior records, a likelihood of the respective consumer engaging in behavior associated with a corresponding one of the psychographic segments, wherein predicting a likelihood of the respective consumer engaging in behavior associated with a corresponding one of the psychographic segments comprises predicting a likelihood to patronize a retail store patronized by others in the same segment.

13. The media of claim 12 , wherein determining more than 5 psychographic segments of consumers comprises:

clustering the first set of consumer-behavior vectors based on pairwise distance in a parameter space defined by attributes of the consumer-behavior records.

14. The media of claim 13 , wherein clustering the first set of consumer behavior vectors comprises determining that at least some vectors have more than a threshold number of other vectors within a threshold distance in the parameter space.

15. The media of claim 14 , wherein determining more than 5 psychographic segments of consumers comprises:

selecting N cluster centers for N candidate clusters from among the first set of vectors, where N is an integer greater than or equal to 5;

assigning vectors in the first set to the candidate clusters based on distance from the respective cluster center; and

iteratively updating the cluster centers to reduce an aggregate measure of distances from the vectors to cluster centers and re-assigning the vectors to updated clusters until less than a threshold amount of vectors change clusters between iterations.

16. The media of claim 12 , wherein determining more than 5 psychographic segments of consumers comprises:

determining at least one segment with at least one latent variable model based on a mixture model of the first set of consumer-behavior vectors.

17. The media of claim 12 , wherein determining more than 5 psychographic segments of consumers comprises:

forming a 2 or higher dimensional lattice of nodes, each node having a node vector and a coordinate in the lattice, the node vector having the same number of dimensions as the vectors in the first set of consumer-behavior records;

initializing the node vectors;

selecting a given consumer-behavior vector from among the first set of consumer-behavior vectors;

for each of at least half the nodes, determining a matching node having an associated node vector that is most similar to the given consumer-behavior vector;

determining adjacent nodes within a threshold distance in the lattice; and

adjusting node vectors of the adjacent nodes to be more similar to the given consumer-behavior vector.

18. The media of claim 12 , wherein:

determining more than 5 psychographic segments of consumers by training an unsupervised machine learning model based on the consumer-behavior vectors comprises:

determining at least 5 corresponding envelopes in a parameter space of the first set of consumer behavior records;

classifying the second set of consumer-behavior vectors as each belonging to at least a respective one of psychographic segments with the trained machine learning model comprises:

determining which parameter space envelope contains the respective member of the second set of consumer-behavior vectors.

19. The media of claim 12 , wherein predicting a likelihood of the respective consumer engaging in behavior associated with a corresponding one of the psychographic segments comprises:

predicting a likelihood to transition from patronizing one plurality of retail businesses to patronizing another plurality of retail businesses transitioned to by others in the same segment.

20. The media of claim 12 , wherein determining a property associated with a corresponding one of the psychographic segments and not associated with the respective consumer comprises:

identifying one or more properties associated with the corresponding one of the psychographic segments and not indicated in the consumer-behavior vector of the respective consumer.

21. The media of claim 12 , wherein determining a property associated with a corresponding one of the psychographic segments and not associated with the respective consumer comprises:

determining a set of properties associated with the corresponding one of the psychographic segments that are not represented in the consumer-behavior vector of the respective consumer.

22. The media of claim 12 , wherein determining a property associated with a corresponding one of the psychographic segments and not associated with the respective consumer comprises:

ranking a plurality of properties according to an amount or frequency of representation of each property within the corresponding one of the psychographic segments; and

selecting the property for the respective consumer based on respective rankings of the plurality of properties.

23. The media of claim 12 , wherein selecting the property for the respective consumer based on respective rankings of the plurality of properties further comprises:

selecting the property based on the consumer-behavior vector of the respective consumer not exhibiting the property.

24. The media of claim 12 , further comprising:

transmitting, to a computing device associated with the respective consumer, content based on the property.

25. The media of claim 12 , wherein determining more than 5 psychographic segments of consumers by training an unsupervised machine learning model based on the consumer-behavior vectors comprises:

establishing each consumer-behavior vector in the first set of consumer-behavior vectors within a parameter space;

selecting a given consumer-behavior vector as a core vector if at least a threshold number of the other computer-behavior vectors are within a threshold distance in the parameter space;

iterating through each of the consumer-behavior vectors to form clusters of consumer-behavior vectors, wherein a cluster is formed based on:

non-core vectors being within a threshold distance of a core vector, and

core vectors being reachable by other core vectors, wherein a first core vector is determined as being reachable from a second core vector based on identification a path of linking core vectors within a threshold distance of one another; and

defining at least one segment based in part on a respective cluster of consumer-behavior vectors.

26. The media of claim 25 , further comprising:

adjusting one or more parameters by which core vectors are selected.

27. The media of claim 15 , further comprising:

adjusting one or more parameters by which cluster centers are selected;

selecting N new cluster centers for N new candidate clusters from among the first set of vectors;

assigning vectors in the first set to the new candidate clusters based on distance from the respective new cluster center;

iteratively updating the new cluster centers to reduce an aggregate measure of distances from the vectors to cluster centers and re-assigning the vectors to updated new clusters until less than a threshold amount of vectors change clusters between iterations; and

selecting either the updated clusters or the updated new clusters based on respective measures of distances from vectors to cluster centers.

28. The media of claim 27 , wherein adjusting one or more parameters by which cluster centers are selected comprises:

randomizing one or more parameters by which cluster centers are initially selected.

29. The media of claim 12 , wherein determining, with one or more computers, more than 5 psychographic segments of consumers by training an unsupervised machine learning model with the first set of consumer-behavior vectors comprises:

defining the consumer-behavior vectors within a parameter space based on attributes of the respective consumer-behavior records;

determining at least one threshold boundary for probabilistically classifying a vector as a member of given psychographic segment; and

determining, based on the at least one threshold boundary and the vector, a probability that the vector is a member of the given psychographic segment.

30. The media of claim 12 , wherein training the unsupervised machine learning model comprises:

determining the psychographic segments without defining a plurality of attributes of the respective psychographic segments in advance of the training.

31. The media of claim 12 , wherein training the unsupervised machine learning model based on the first set of consumer-behavior vectors comprises:

training on a plurality of attributes associated with respective ones of the consumer-behavior vectors in the first set of consumer-behavior vectors that are not pre-labeled.

32. The media of claim 12 , wherein determining more than 5 psychographic segments of consumers by training an unsupervised machine learning model comprises:

training the unsupervised machine learning model on a plurality of attributes associated with respective ones of the consumer-behavior vectors in the first set of consumer-behavior vectors that are not pre-labeled, and

wherein the psychographic segments are determined by training the unsupervised machine learning model without defining a plurality of attributes of the respective psychographic segments in advance of the training.

33. The media of claim 12 , wherein converting the first set of consumer-behavior records into respective consumer-behavior vectors in a first set of consumer-behavior vectors comprises:

identifying a plurality of records as being associated with a given consumer based on a matching of identifying information indicative of the given consumer across the plurality of records;

forming a consumer-behavior record of the given consumer based on the plurality of records; and

converting the consumer-behavior record of the given consumer into a consumer-behavior vector encoding a plurality of attributes of the given consumer based on the plurality of records.

34. The media of claim 33 , wherein:

each of the plurality of records comprises a respective timestamp, and forming the consumer-behavior record of the given consumer based on the plurality of records comprises a selection of records from the plurality of records having respective timestamps covering a trailing duration of time.

35. The media of claim 33 , wherein:

each of the plurality of records corresponds to a respective indication of a discrete instance of behavior of the given consumer, and

the consumer-behavior record comprises multiple instances of behavior of the given consumer.

36. A non-transitory, machine-readable media storing instructions that when executed by one or more computers effectuate operations comprising:

obtaining, with one or more computers, a first set of more than 10,000 consumer-behavior records, each consumer-behavior record indicating, for a respective consumer, at least two businesses patronized by the respective consumer;

converting, with one or more computers, the first set of consumer-behavior records into respective consumer-behavior vectors in a first set of consumer-behavior vectors;

determining, with one or more computers, more than 5 psychographic segments of consumers by training an unsupervised machine learning model with the first set of consumer-behavior vectors, wherein the first set of consumer-behavior vectors are not pre-labeled as members of any of the psychographic segments before determining the psychographic segments;

obtaining, with one or more computers, a second set of consumer-behavior records after determining the psychographic segments of consumers;

converting, with one or more computers, the second set of consumer-behavior records into respective consumer-behavior vectors in a second set of consumer-behavior vectors;

classifying, with one or more computers, the second set of consumer-behavior vectors as each belonging to at least a respective one of psychographic segments with the trained machine learning model;

determining, with one or more computers, based on the classification, for at least some of the second set of consumer behavior records, a property associated with a corresponding one of the psychographic segments and not associated with the respective consumer; and

predicting, with one or more computers, based on the classification, for each of the second set of consumer behavior records, a likelihood of the respective consumer engaging in behavior associated with a corresponding one of the psychographic segments, wherein predicting a likelihood of the respective consumer engaging in behavior associated with a corresponding one of the psychographic segments comprises predicting a likelihood to transition from patronizing one plurality of retail businesses to patronizing another plurality of retail businesses transitioned to by others in the same segment.

Assignments (5)

Continuity (18)