System and method using deep learning machine vision to conduct product positioning analyses
At least one embodiment is directed to a computer-implemented method for using machine vision to categorize a locality to conduct product positioning analyses, the method including: generating locality profile scores for each locality of a plurality of localities using deep learning networks, where the locality profile score includes distributions of entity classes within the locality; extracting a set of entities having the same entity class from a group of localities; retrieving historical purchasing data for the entity set; and generating a sequence of products likely to be purchased by a target entity as a function of: the similarity of purchasing characteristics of the target entity with respect to other entities, product sequences found in product purchase of other entities, and entity profile weights extracted from the locality profile scores of other entities that have purchased one or more of the same products as the target entity.
1. A computer-implemented method for using machine vision to categorize a locality to conduct product positioning analyses, the method comprising:
training a neural network to extract regions of segmented pixel areas of a map image to provide a trained neural network, the neural network comprising a first convolutional neural network and a second convolutional network, the training comprising classifying entities from a plurality of entities based upon at least one of text and icons represented within the segmented pixel areas, the classifying entities classifying entities into a first type and a second type, the first convolutional neural network being trained using entities of the first type and the second convolutional neural network being trained using entities of the second type, the training the neural network enabling deep learning machine vision analysis on geographic artefacts found in the map image;
generating locality profile scores for each locality of a plurality of localities, wherein the locality profile score includes distributions of entity classes within the locality, the locality profile score for each locality being derived through neural network analyses of map images of the locality, the neural network analysis using the trained neural network;
extracting a set of entities having the same entity class from a group of localities, the set of entities being extracted using the trained neural network;
retrieving historical purchasing data for the set of entities for products purchased over a predetermined period of time;
determining similarity of purchasing characteristics for a target entity in the set of entities with respect to other entities in the set of entities;
generating a sequence of products likely to be purchased by the target entity as a function of:
similarity of purchasing characteristics of the target entity with respect to other entities in the set of entities,
product sequences found in product purchase histories in the historical purchasing data of other entities in the set of entities, and
entity profile weights extracted from the locality profile scores of other entities in the set of entities that have purchased one or more of the same products as the target entity;
assigning purchases of products within the predetermined period of time to respective sub-periods of time for each entity in the set of entities, wherein the sub-periods of time are equally divided into sub-periods of length M, where M-x corresponds to a sub-period of time having a distance index −x from an endpoint of the predetermined period of time; and,
for each entity in the set of entities, predicting a subsequent number “n” of products that will likely be purchased by the entity using operations including:
identifying a most recent product (MRP) purchased by the target entity within the predetermined period of time;
searching for purchases of the MRP by other entities in the historical purchasing data;
if a purchase of the MRP is found for another entity,
identifying a position M-x of the sub-period of time in which the product was purchased, and
identifying an “n” sequence of products purchased by the other entity; and
generating a product recommendation score for each product of the “n” product sequence for the other entity, wherein the product recommendation score is a function of the similarity of product purchasing characteristics of the target entity and other entity, an entity profile weight of the other entity, and the position of the product in the “n” product sequence purchased by the other entity.
2. The computer-implemented method of claim 1 , wherein identifying entities having similar product purchasing patterns comprises:
determining a standardized monthly revenue for each type of product purchased by each entity in the set of entities over the predetermined period of time; and
comparing the standardized revenue of products purchased by a target entity over the predetermined period of time with the standardized revenue of products purchased by other entities in the set of entities to calculate similarity scores indicative of similarities of purchasing characteristics of the target entity with respect to other entities in the set of entities.
3. The computer-implemented method of claim 2 , wherein
the standardized monthly revenue for a type of a product purchased by an entity in the set of entities includes determining a proportion of spending on the product with respect to spending by the entity on other types of products purchased by the entity during the predetermined period of time; and
the similarity scores are calculated using a cosine similarity operation.
4. The computer-implemented method of claim 1 , further comprising:
generating product recommendation scores for the products of the “n” product sequences; and
using the highest “n” product recommendation scores to identify the “n” product sequence of products likely to be purchased by the target entity.
5. The computer-implemented method of claim 1 , further comprising:
identifying greenfield entities through neural network analyses of the map images of localities in the group of localities;
obtaining firmographics information for the greenfield entities; and
generating a sequence of products likely to be purchased by the greenfield entities using the product sequences generated for one or more entities in the set of entities having similar firmographics as the greenfield entities.
6. A system comprising:
a processor;
a data bus coupled to the processor; and
a non-transitory, computer-readable storage medium embodying computer program code, the non-transitory, computer-readable storage medium being coupled to the data bus, the computer program code interacting with a plurality of computer operations and comprising instructions executable by the processor and configured for:
training a neural network to extract regions of segmented pixel areas of a map image to provide a trained neural network, the neural network comprising a first convolutional neural network and a second convolutional network, the training comprising classifying entities from a plurality of entities based upon at least one of text and icons represented within the segmented pixel areas, the classifying entities classifying entities into a first type and a second type, the first convolutional neural network being trained using entities of the first type and the second convolutional neural network being trained using entities of the second type, the training the neural network enabling deep learning machine vision analysis on geographic artefacts found in the map image;
generating locality profile scores for each locality of a plurality of localities, wherein the locality profile score includes distributions of entity classes within the locality, the locality profile score for each locality being derived through neural network analyses of map images of the locality, the neural network analysis using the trained neural network;
extracting a set of entities having the same entity class from a group of localities, the set of entities being extracted using the trained neural network;
retrieving historical purchasing data for the set of entities for products purchased over a predetermined period of time;
determining similarity of purchasing characteristics for a target entity in the set of entities with respect to other entities in the set of entities;
generating a sequence of products likely to be purchased by the target entity as a function of:
similarity of purchasing characteristics of the target entity with respect to other entities in the set of entities,
product sequences found in product purchase histories in the historical purchasing data of other entities in the set of entities, and
entity profile weights extracted from the locality profile scores of other entities in the set of entities that have purchased one or more of the same products as the target entity;
assigning purchases of products within the predetermined period of time to respective sub-periods of time for each entity in the set of entities, wherein the sub-periods of time are equally divided into sub-periods of length M, where M-x corresponds to a sub-period of time having a distance index −x from an endpoint of the predetermined period of time; and,
for each entity in the set of entities, predicting a subsequent number “n” of products that will likely be purchased by the entity using operations including:
identifying a most recent product (MRP) purchased by the target entity within the predetermined period of time;
searching for purchases of the MRP by other entities in the historical purchasing data;
if a purchase of the MRP is found for another entity,
identifying a position M-x of the sub-period of time in which the product was purchased, and
identifying an “n” sequence of products purchased by the other entity; and
generating a product recommendation score for each product of the “n” product sequence for the other entity, wherein the product recommendation score is a function of the similarity of product purchasing characteristics of the target entity and other entity, an entity profile weight of the other entity, and the position of the product in the “n” product sequence purchased by the other entity.
7. The system of claim 6 , wherein identifying entities having similar product purchasing characteristics comprises:
determining a standardized monthly revenue for each type of product purchased by each entity in the set of entities over the predetermined period of time; and
comparing the standardized revenue of products purchased by a target entity over the predetermined period of time with the standardized revenue of products purchased by other entities in the set of entities to calculate similarity scores indicative of similarities of purchasing characteristics of the target entity with respect to other entities in the set of entities.
8. The system of claim 7 , wherein
the standardized monthly revenue for a type of a product purchased by an entity in the set of entities includes determining a proportion of spending on the product with respect to spending by the entity on other types of products purchased by the entity during the predetermined period of time; and
similarity scores are calculated using a cosine similarity operation.
9. The system of claim 6 , further comprising:
generating product recommendation scores for the products of the “n” product sequences; and
using the highest “n” product recommendation scores to identify the “n” product sequence of products likely to be purchased by the target entity.
10. The system of claim 6 , further comprising:
identifying greenfield entities through neural network analyses of the map images of localities in the group of localities;
obtaining firmographics information for the greenfield entities; and
generating a sequence of products likely to be purchased by the greenfield entities using the product sequences generated for one or more entities in the set of entities having similar firmographics as the greenfield entities.
11. A non-transitory, computer-readable storage medium embodying computer program code, the computer program code comprising computer executable instructions configured for:
training a neural network to extract regions of segmented pixel areas of a map image to provide a trained neural network, the neural network comprising a first convolutional neural network and a second convolutional network, the training comprising classifying entities from a plurality of entities based upon at least one of text and icons represented within the segmented pixel areas, the classifying entities classifying entities into a first type and a second type, the first convolutional neural network being trained using entities of the first type and the second convolutional neural network being trained using entities of the second type, the training the neural network enabling deep learning machine vision analysis on geographic artefacts found in the map image;
generating locality profile scores for each locality of a plurality of localities, wherein the locality profile score includes distributions of entity classes within the locality, the locality profile score for each locality being derived through neural network analyses of map images of the locality, the neural network analysis using the trained neural network;
extracting a set of entities having the same entity class from a group of localities, the set of entities being extracted using the trained neural network;
retrieving historical purchasing data for the set of entities for products purchased over a predetermined period of time;
determining similarity of purchasing characteristics for a target entity in the set of entities with respect to other entities in the set of entities; and
generating a sequence of products likely to be purchased by the target entity as a function of:
similarity of purchasing characteristics of the target entity with respect to other entities in the set of entities,
product sequences found in product purchase histories in the historical purchasing data of other entities in the set of entities, and
entity profile weights extracted from the locality profile scores of other entities in the set of entities that have purchased one or more of the same products as the target entity;
assigning purchases of products within the predetermined period of time to respective sub-periods of time for each entity in the set of entities, wherein the sub-periods of time are equally divided into sub-periods of length M, where M-x corresponds to a sub-period of time having a distance index −x from an endpoint of the predetermined period of time; and,
for each entity in the set of entities, predicting a subsequent number “n” of products that will likely be purchased by the entity using operations including:
identifying a most recent product (MRP) purchased by the target entity within the predetermined period of time;
searching for purchases of the MRP by other entities in the historical purchasing data;
if a purchase of the MRP is found for another entity,
identifying a position M-x of the sub-period of time in which the product was purchased, and
identifying an “n” sequence of products purchased by the other entity; and
generating a product recommendation score for each product of the “n” product sequence for the other entity, wherein the product recommendation score is a function of the similarity of product purchasing characteristics of the target entity and other entity, an entity profile weight of the other entity, and the position of the product in the “n” product sequence purchased by the other entity.
12. The non-transitory, computer-readable storage medium of claim 11 , wherein
identifying entities having similar product purchasing characteristics comprises:
determining a standardized monthly revenue for each type of product purchased by each entity in the set of entities over the predetermined period of time; and
comparing the standardized revenue of products purchased by a target entity over the predetermined period of time with the standardized revenue of products purchased by other entities in the set of entities to calculate similarity scores indicative of similarities of purchasing characteristics of the target entity with respect to other entities in the set of entities.
13. The non-transitory, computer-readable storage medium of claim 12 , wherein
the standardized monthly revenue for a type of a product purchased by an entity in the set of entities includes determining a proportion of spending on the product with respect to spending by the entity on other types of products purchased by the entity during the predetermined period of time; and
the similarity scores are calculated using a cosine similarity operation.
14. The non-transitory, computer-readable storage medium of claim 11 , further comprising:
generating product recommendation scores for the products of the “n” product sequences; and
using the highest “n” product recommendation scores to identify the “n” product sequence of products likely to be purchased by the target entity.