U.S. patent application number 16/539608 was filed with the patent office on 2020-02-20 for system and method for determination of export codes.
The applicant listed for this patent is Walmart Apollo, LLC. Invention is credited to Jevon T. Jamieson, Stewart D. Lewis, Alyssa J. Schrade, Ahalya Vikram.
Application Number | 20200057987 16/539608 |
Document ID | / |
Family ID | 69522961 |
Filed Date | 2020-02-20 |
United States Patent
Application |
20200057987 |
Kind Code |
A1 |
Schrade; Alyssa J. ; et
al. |
February 20, 2020 |
SYSTEM AND METHOD FOR DETERMINATION OF EXPORT CODES
Abstract
The export of products from a jurisdiction is enabled by
choosing export codes from a model. The electronic model comprises
a branched tree defining a configuration of the export codes and
selection criteria for the codes. The branched tree includes a
plurality of branches and leaves. Each of the leaves includes an
export code for a product, a probability that the export code for
the product is accurate, and a description of the product. The tree
is traversed to obtain the codes. The tree is trained to improve
the quality of the selection process.
Inventors: |
Schrade; Alyssa J.; (Cave
Springs, AR) ; Lewis; Stewart D.; (Centerton, AR)
; Jamieson; Jevon T.; (Rogers, AR) ; Vikram;
Ahalya; (Bentonville, AR) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
Walmart Apollo, LLC |
Bentonville |
AR |
US |
|
|
Family ID: |
69522961 |
Appl. No.: |
16/539608 |
Filed: |
August 13, 2019 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
62719845 |
Aug 20, 2018 |
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06Q 10/0831 20130101;
G06F 16/9035 20190101; G06F 16/951 20190101; G06N 20/00 20190101;
G06Q 10/0838 20130101; G06N 3/0454 20130101; G06N 5/003 20130101;
G06N 3/08 20130101 |
International
Class: |
G06Q 10/08 20060101
G06Q010/08; G06N 20/00 20060101 G06N020/00 |
Claims
1. A system for enabling the export of products from a
jurisdiction, the system comprising: an electronic communication
network; an electronic database that stores a product information
table of currently classified products, each entry of the table
including a product number and at least one attribute of a
currently classified product; an electronic model stored in the
database, the electronic model comprising a branched tree defining
a configuration of export codes and selection criteria for the
codes, the branched tree including a plurality of branches and
leaves, each of the leaves including an export code for a product,
a probability that the export code for the product is accurate, and
a description of the product; a control circuit, the control
circuit being coupled to the electronic communication network and
the database, the control circuit configured to: via the electronic
communication network, web-scrape external web sites for product
information concerning products presently represented in the model
and similar products not presently represented in the model; create
synthetic export codes for selected products not presently
represented in the model according to machine learning approaches
using the web-scraped information and export codes obtained from a
regulatory source of a jurisdiction; adjust the model based at
least in part upon one or more of the classified product
information from the database, the web-scraped product information,
and the synthetic export codes; wherein the system further
includes: a shipping vehicle; one or more products-to-be-shipped
outside the jurisdiction, the products-to-be-shipped using the
shipping vehicle, the products requiring export code assignment
before the vehicle leaves the jurisdiction; an electronic user
device coupled to the network; and one or more electronic sensors
coupled to the electronic user device, the sensors being configured
to obtain an image of one of the products to-be-shipped, wherein
the electronic user device upon user action forms a request to
export the products, the request including the image; wherein the
control circuit is further configured to: receive the request;
based upon an analysis of the image, traverse the branched tree of
the model to obtain one or more potential export codes for the
products-to-be-shipped, the potential export codes being situated
at one or more leaves of the tree and each of the potential export
codes having an associated probability; when the associated
probability associated with one of the potential export codes is
greater than a predetermined threshold, select the one potential
export code; when probability of all the potential export codes is
less than a threshold, then select one of the potential export
codes based upon a predetermined criteria; transmit the selected
export code to an electronic processing device of the regulatory
source via the electronic communication network; wherein the
shipping vehicle responsively exits the jurisdiction with the
products only after the selected export code is determined.
2. The system of claim 1, wherein the request includes additional
product information obtained from a user at the electronic user
device via the user interface.
3. The system of claim 1, wherein the adjustment is made
periodically at regular intervals.
4. The system of claim 1, wherein the sensors include cameras.
5. The system of claim 1, wherein the jurisdiction is a
country.
6. The system of claim 1, wherein the shipping vehicle is a truck,
train, ship, automated ground vehicle, or aerial drone.
7. The system of claim 1, wherein the control circuit and database
are disposed at a central processing center.
8. The system of claim 1, wherein the predetermined criteria is
automatically determined based upon historical data.
9. The system of claim 1, wherein the predetermined criteria is
determined by a human operator.
10. A method for enabling the export of products from a
jurisdiction, the method comprising: providing an electronic
communication network; providing an electronic database that stores
a product information table of currently classified products, each
entry of the table including a product number and at least one
attribute of a currently classified product; storing an electronic
model in the database, the electronic model comprising a branched
tree defining a configuration of export codes and selection
criteria for the codes, the branched tree including a plurality of
branches and leaves, each of the leaves including an export code
for a product, a probability that the export code for the product
is accurate, and a description of the product; at a control circuit
and via an electronic communication network, web-scraping external
websites for product information concerning products presently
represented in the model and similar products not presently
represented in the model; at the control circuit, creating
synthetic export codes for selected products not presently
represented in the model according to machine learning approaches
using the web-scraped information and export codes obtained from a
regulatory source of a jurisdiction; at the control circuit,
adjusting the model based at least in part upon one or more of the
classified product information from the database, the web-scraped
product information, and the synthetic export codes; providing a
shipping vehicle and one or more products-to-be-shipped outside the
jurisdiction, the products-to-be-shipped using the shipping
vehicle, the products requiring export code assignment before the
vehicle leaves the jurisdiction; providing an electronic user
device coupled to the network; providing one or more electronic
sensors coupled to the electronic user device, the sensors being
configured to obtain an image of one of the products to-be-shipped,
wherein the electronic user device upon user action forms a request
to export the products, the request including the image; receive
the request at the control circuit; based upon an analysis of the
image, traversing the branched tree of the model to obtain one or
more potential export codes for the products-to-be-shipped, the
potential export codes being situated at one or more leaves of the
tree and each of the potential export codes having an associated
probability; at the control circuit and when the associated
probability associated with one of the potential export codes is
greater than a predetermined threshold, selecting the one potential
export code; at the control circuit and when probability of all the
potential export codes is less than a threshold, then select one of
the potential export codes based upon a predetermined criteria; at
the control circuit, transmitting the selected export code to an
electronic processing device of the regulatory source via the
electronic communication network; wherein the shipping vehicle
responsively exits the jurisdiction with the products only after
the selected export code is determined.
11. The method of claim 10, wherein the request includes additional
product information obtained from a user at the electronic user
device via a user interface.
12. The method of claim 10, wherein the adjustment is made
periodically at regular intervals.
13. The method of claim 10, wherein the sensors include
cameras.
14. The method of claim 10, wherein the jurisdiction is a
country.
15. The method of claim 10, wherein the shipping vehicle is a
truck, train, ship, automated ground vehicle, or aerial drone.
16. The method of claim 10, wherein the control circuit and
database are disposed at a central processing center.
17. The method of claim 10, wherein the predetermined criteria is
automatically determined based upon historical data.
18. The method of claim 10, wherein the predetermined criteria is
determined by a human operator.
Description
CROSS-REFERENCE TO RELATED APPLICATION
[0001] This application claims the benefit of the following U.S.
Provisional Application No. 62/719,845 filed Aug. 20, 2018, which
is incorporated herein by reference in its entirety.
TECHNICAL FIELD
[0002] These teachings relate to export codes or information and,
more specifically, to the determination of this information.
BACKGROUND
[0003] Products are shipped across the world from and between
different locations and jurisdictions. When leaving a country, for
example, the United States, an export code needs to be obtained.
The export code is associated with a particular product and
typically is associated with details concerning that product. For
example, an export code may be associated with men's cotton
t-shirts. Other examples of export codes are possible.
[0004] In the United States, the Harmonized Tariff Schedule (also
known as Schedule B) is the primary source of determining the
tariff (custom duties) for goods exported from the United States to
foreign countries. The Harmonized Tariff Schedule classifies a good
or product based upon its name, use and/or material used in its
construction and assigns it a ten-digit classification code number.
There are over 9,000 classification code numbers.
[0005] Export codes can be used for various purposes. In one
particular example, the export codes are used to track the nature
and amount of products leaving a jurisdiction.
BRIEF DESCRIPTION OF THE DRAWINGS
[0006] The above needs are at least partially met through the
provision of approaches that determine action to take upon a
product recall, wherein:
[0007] FIG. 1 comprises a diagram of a system as configured in
accordance with various embodiments of these teachings;
[0008] FIG. 2 comprises a flowchart as configured in accordance
with various embodiments of these teachings;
[0009] FIG. 3 comprises a flowchart as configured in accordance
with various embodiments of these teachings;
[0010] FIG. 4 comprises a flowchart as configured in accordance
with various embodiments of these teachings;
[0011] FIG. 5 comprises a flowchart as configured in accordance
with various embodiments of these teachings;
[0012] FIG. 6 comprises a flowchart as configured in accordance
with various embodiments of these teachings;
[0013] FIG. 7 comprises a flowchart as configured in accordance
with various embodiments of these teachings;
[0014] FIG. 8 comprises a diagram of a data structure as configured
in accordance with various embodiments of these teachings;
DETAILED DESCRIPTION
[0015] Generally speaking, in the present approaches, an electronic
model defines a configuration of export codes and rules to choose
these codes. In aspects, the model is a tree-like data structure
and when the model is traversed according to product information
for a product, the export code is determined for the product. The
model is periodically fine-tuned and improved to include or
incorporate web-scraped information and information from other
sources. When a new product is to be exported and needs a code,
product images and other information are obtained for the product
and the images are applied to the model to obtain the export code.
Once the code is obtained, the product can be exported on a
delivery vehicle and this information can also be provided to a
government authority that monitors exports.
[0016] In the approaches described here, the model learns from
various types of data, e.g., previously classified items,
Harmonized Tariff Schedule (HTS) codes, product descriptions,
product images, and web-scraped information to mention a few
examples. The model is continuously (or periodically) refined and
updated. The model can then be used to determine one of the unique
HTS codes using, for example, the images and product description
given by suppliers of the product (or others).
[0017] In these regards, information can be obtained from item
information databases, online catalogs, and the internet to mention
a few examples. Each product or good typically requires a certain
type of information to classify (and obtain an export code). For
example, shoes may need the length of the ankle, material of the
upper part of the shoe, and the material of the lower part of the
shoe. Food products may require the ingredients found in the
products. When this information is not available from some sources
(e.g., from the supplier or manufacturer), it may be obtained from
other sources (e.g., by web-scraping information from websites on
the internet).
[0018] After gathering the item description data (from whatever
source), a pre-process can be utilized to clean the information to
determine, for example, the name, use, and material used to
determine the code. Cleaning may include obtaining only the
required data for a given product or commodity, removing
information such as brand name and product title, correcting
spelling errors, understanding abbreviations, or splitting words
that have become erroneously attached to each other.
[0019] Not all of the 9000+ HTS codes have examples that have been
classified. Consequently, the model is trained to include new HTS
codes that are synthesized. Historical or other information can be
used to train the model. In other examples, images obtained from
on-line catalog are used to train the last layers of a pretrained
model in a process known as transfer learning to detect the edges
and shapes in the images.
[0020] Using the model, items can be classified using both text
(describing the product or item) and images (taken of the product
or item). The output of this process is an export code, for
example, a HTS code.
[0021] In many of these embodiments, a system for enabling the
export of products from a jurisdiction includes an electronic
communication network, an electronic database, an electronic model,
a control circuit, a shipping vehicle, one or more
products-to-be-shipped, an electronic user device, and one or more
electronic sensors.
[0022] The electronic database stores a product information table
of currently classified products. Each entry of the table includes
a product number and at least one attribute of a currently
classified product.
[0023] The electronic model is stored in the database and comprises
a branched tree defining a configuration of export codes and
selection criteria for the codes. The branched tree includes a
plurality of branches and leaves, and each of the leaves includes
an export code for a product, a probability that the export code
for the product is accurate, and a description of the product.
[0024] The control circuit is coupled to the electronic
communication network and the database. The control circuit is
configured to, via the electronic communication network, web-scrape
external websites for product information concerning products
presently represented in the model and similar products not
presently represented in the model. The control circuit is further
configured to create synthetic export codes for selected products
not presently represented in the model according to machine
learning approaches using the web-scraped information and export
codes obtained from a regulatory source of a jurisdiction. The
control circuit is additionally configured to adjust the model
based at least in part upon one or more of the classified product
information from the database, the web-scraped product information,
and the synthetic export codes.
[0025] The one or more products-to-be-shipped outside the
jurisdiction are shipped using the shipping vehicle. The products
requiring export code assignment before the vehicle leaves the
jurisdiction. The electronic user device coupled to the
network.
[0026] The one or more electronic sensors are coupled to the
electronic user device. The sensors are configured to obtain an
image of one of the products to-be-shipped. The electronic user
device, upon user action, forms a request to export the products,
the request including the image.
[0027] The control circuit is further configured to receive the
request. The control circuit is configured to, based upon an
analysis of the image, traverse the branched tree of the model to
obtain one or more potential export codes for the
products-to-be-shipped. The potential export codes are situated at
one or more leaves of the tree and each of the potential export
codes having an associated probability. When the associated
probability associated with one of the potential export codes is
greater than a predetermined threshold, the one potential export
code is selected. When probability of all the potential export
codes is less than a threshold, then one of the potential export
codes is selected based upon a predetermined criteria.
[0028] The selected export code is transmitted to an electronic
processing device of the regulatory source via the electronic
communication network. The shipping vehicle responsively exits the
jurisdiction with the products only after the selected export code
is determined.
[0029] In aspects, the request includes additional product
information obtained from a user at the electronic user device via
a user interface at the electronic device. In other aspects, the
adjustment is made periodically at regular intervals.
[0030] In other examples, the sensors include cameras. In still
other examples, the jurisdiction is a country.
[0031] In other aspects, the delivery vehicle is a truck, train,
ship, automated ground vehicle, or aerial drone. Other examples are
possible.
[0032] In other examples, the control circuit and database are
disposed at a central processing center. In yet other examples, the
predetermined criteria is automatically determined based upon
historical data. In some examples, the predetermined criteria is
determined by a human operator.
[0033] In others of these embodiments, the export of products from
a jurisdiction is enabled. An electronic communication network is
provided. An electronic database that stores a product information
table of currently classified products is also provided. Each entry
of the table includes a product number and at least one attribute
of a currently classified product.
[0034] An electronic model is stored in the database and the
electronic model comprises a branched tree defining a configuration
of export codes and selection criteria for the codes. The branched
tree includes a plurality of branches and leaves. Each of the
leaves includes an export code for a product, a probability that
the export code for the product is accurate, and a description of
the product.
[0035] At a control circuit and via an electronic communication
network, external websites are web-scraped for product information
concerning products presently represented in the model and similar
products not presently represented in the model. At the control
circuit, synthetic export codes are created for selected products
not presently represented in the model according to machine
learning approaches using the web-scraped information and export
codes obtained from a regulatory source of a jurisdiction.
[0036] At the control circuit, the model is adjusted based at least
in part upon one or more of the classified product information from
the database, the web-scraped product information, and the
synthetic export codes. A shipping vehicle and one or more
products-to-be-shipped outside the jurisdiction are provided. The
products-to-be-shipped use the shipping vehicle and the products
requiring export code assignment before the vehicle leaves the
jurisdiction.
[0037] An electronic user device is coupled to the network. One or
more electronic sensors are coupled to the electronic user device,
and the sensors are configured to obtain an image of one of the
products to-be-shipped. The electronic user device, upon user
action, forms a request to export the products, and the request
includes the image.
[0038] The request at the control circuit. Based upon an analysis
of the image, the branched tree of the model is traversed to obtain
one or more potential export codes for the products-to-be-shipped.
The potential export codes are situated at one or more leaves of
the tree and each of the potential export codes has an associated
probability.
[0039] At the control circuit and when the associated probability
associated with one of the potential export codes is greater than a
predetermined threshold, the one potential export code is selected.
At the control circuit and when probability of all the potential
export codes is less than a threshold, then one of the potential
export codes is selected based upon a predetermined criteria.
[0040] At the control circuit, the selected export code is
transmitted to an electronic processing device of the regulatory
source via the electronic communication network. The shipping
vehicle responsively exits the jurisdiction with the products only
after the selected export code is determined.
[0041] Referring now to FIG. 1, a system 100 for enabling the
export of products from a jurisdiction is described. The approaches
described herein result in obtaining export codes, which may
include letters, numbers, and/or special characters. However, it
will be appreciated that these approaches may be utilized to
determine other types of classification information and that the
codes and/or information need not be strictly related to the export
of products or goods. The system 100 includes an electronic
communication network 102, an electronic database 104, an
electronic model 106, a control circuit 108, a shipping vehicle
110, one or more products-to-be-shipped 112, an electronic user
device 114, and one or more electronic sensors 116.
[0042] The electronic communication network 102 is any type of
communication network or combination of networks such as the
internet, cellular communication networks, data networks, wide area
networks, local area networks, wireless networks, or any other type
of electronic communication network.
[0043] The electronic database 104 is any type of electronic memory
storage device. The electronic database 104 stores a product
information table of currently classified products. Each entry of
the table includes a product number and at least one attribute of a
currently classified product.
[0044] The electronic model 106 is stored in the database and the
electronic model comprises a branched tree defining a configuration
of export codes and selection criteria for the codes. The branched
tree includes a plurality of branches and leaves. Each of the
leaves includes an export code for a product, a probability that
the export code for the product is accurate, and a description of
the product.
[0045] The control circuit 108 is coupled to the electronic
communication network 102 and the database 104. It will be
appreciated that as used herein the term "control circuit" refers
broadly to any microcontroller, computer, or processor-based device
with processor, memory, and programmable input/output peripherals,
which is generally designed to govern the operation of other
components and devices. It is further understood to include common
accompanying accessory devices, including memory, transceivers for
communication with other components and devices, etc. These
architectural options are well known and understood in the art and
require no further description here. The control circuit 108 may be
configured (for example, by using corresponding programming stored
in a memory as will be well understood by those skilled in the art)
to carry out one or more of the steps, actions, and/or functions
described herein. In other examples, the control circuit and
database are disposed at a central processing center.
[0046] The shipping vehicle 110 is a truck, train, ship, automated
ground vehicle, or aerial drone. Other examples are possible.
[0047] The products-to-be-shipped 112 outside the jurisdiction are
shipped using the shipping vehicle. The products requiring export
code assignment before the vehicle 110 leaves the jurisdiction. In
still other examples, the jurisdiction is a country.
[0048] The electronic user device 114 is any type of electronic
device such as a personal computer, cellular phone, smartphone,
laptop, or tablet to mention a few examples. The device 114
includes a user interface and is coupled to the network 102.
[0049] The electronic sensors 116. In other examples, the sensors
include cameras. Other examples are possible. The electronic
sensors 116 are coupled to the electronic user device 114.
[0050] The control circuit 108 is configured to, via the electronic
communication network 102, web-scrape external websites 122 for
product information concerning products presently represented in
the model and similar products not presently represented in the
model. The control circuit 108 is further configured to create
synthetic export codes for selected products not presently
represented in the model according to machine learning approaches
using the web-scraped information and export codes 126 obtained
from a regulatory source 128 of a jurisdiction. The control circuit
108 is additionally configured to adjust the model based at least
in part upon one or more of the classified product information 124
from the database, the web-scraped product information, and the
synthetic export codes.
[0051] The sensors 116 are configured to obtain an image of one of
the products to-be-shipped. The electronic user device 114, upon
user action, forms a request to export the products, and the
request includes the image.
[0052] The control circuit 108 is further configured to receive the
request. The control circuit 108 is configured to, based upon an
analysis of the image, traverse the branched tree of the model to
obtain one or more potential export codes for the
products-to-be-shipped 112. The potential export codes are situated
at one or more leaves of the tree and each of the potential export
codes having an associated probability. When the associated
probability associated with one of the potential export codes is
greater than a predetermined threshold, the one potential export
code is selected. When probability of all the potential export
codes is less than a threshold, then one of the potential export
codes is selected based upon a predetermined criteria. In some
examples, the predetermined criteria is automatically determined
based upon historical data. In other examples, the predetermined
criteria is determined by a human operator.
[0053] The selected export code is transmitted to an electronic
processing device 120 of the regulatory source via the electronic
communication network. The shipping vehicle 110 responsively exits
the jurisdiction with the products 112 only after the selected
export code is determined.
[0054] In aspects, the request includes additional product
information obtained from a user at the electronic user device via
a user interface at the electronic device 114. In other aspects,
the adjustment is made periodically at regular intervals.
[0055] Referring now to FIG. 2, an approach for determining the
export code of products to be exported from a jurisdiction is
described. At step 202, an electronic communication network is
provided. The electronic communication network is any type of
communication network or combination of networks such as the
internet, cellular communication networks, data networks, wide area
networks, local area networks, wireless networks, or any other type
of electronic communication network.
[0056] At step 204, an electronic database that stores a product
information table of currently classified products is also
provided. Each entry of the table includes a product number and at
least one attribute of a currently classified product.
Alternatively, other types of data structures can be used.
[0057] At step 206, an electronic model is stored in the database
and the electronic model comprises a branched tree defining a
configuration of export codes and selection criteria for the codes.
The branched tree includes a plurality of branches and leaves. Each
of the leaves includes an export code for a product, a probability
that the export code for the product is accurate, and a description
of the product. The use of the tree data structure allows for the
quick and efficient retrieval of information and is advantageous
compared to other types of data structures that are slower and less
efficient to traverse. Consequently, an underlying control circuit
or computer implementing these approaches itself operates more
quickly and efficiently.
[0058] At step 208, a control circuit (via an electronic
communication network) is used to web-scrape external websites for
product information concerning products presently represented in
the model and similar products not presently represented in the
model. In aspects, a web page is web-scraped by fetching the
web-page and extracting information from the web page. Fetching may
involve downloading of a page from a server or other arrangement.
Once fetched, data or information can be extracted from the web
page. In these regards, the web page may be parsed, searched,
reformatted, or portions copied and stored.
[0059] At step 210 and at the control circuit, synthetic export
codes are created for selected products not presently represented
in the model according to machine learning approaches using the
web-scraped information and export codes obtained from a regulatory
source of a jurisdiction. In one aspect, the codes are generated
based upon an analysis of existing codes. Existing and known codes
(obtained from a database) are analyzed and compared to web-scraped
(or other types of) information. Then, a determination can be made
as to whether to add a new code to the tree, modify a code in the
tree, or delete a code from the tree. For instance, a code may be
located on a government list for woman cotton t-shirts and a
determination made if this classification matches, is confirmed by,
or is not contradicted by web-scraped information. Once confirmed
and if this code is not on the tree, it may be added at an
appropriate location in the tree data structure of the model.
Historical information may be maintained concerning the accuracy of
this code and this historical information also included with the
code on the tree.
[0060] At step 212 and at the control circuit, the model is
adjusted based at least in part upon one or more of the classified
product information from the database, the web-scraped product
information, and the synthetic export codes. As mentioned, this may
involve modifications to leaves of the tree. However, the structure
of the branches of the tree may also be modified (e.g., branches
and relationships in the tree may be modified).
[0061] At step 214, a shipping vehicle and one or more
products-to-be-shipped outside the jurisdiction are provided. The
products-to-be-shipped use the shipping vehicle and the products
requiring export code assignment before the vehicle leaves the
jurisdiction.
[0062] At step 216, an electronic user device is coupled to the
network. The electronic user device may be a personal computer,
cellular phone, smartphone, laptop, or tablet to mention a few
examples.
[0063] At step 218, one or more electronic sensors are coupled to
the electronic user device, and the sensors are configured to
obtain an image of one of the products to-be-shipped. In examples,
the sensors may be cameras. Other examples of sensors are
possible.
[0064] At step 220, the electronic user device, upon user action,
forms a request to export the products, and the request includes
the image. For example, a camera on a smart phone takes a photo of
a product. A user may send an electronic request via email that
includes the photo and potentially other information to the control
circuit.
[0065] At step 222, the request is received at the control circuit.
The control circuit analyzes the image, for example, using image
recognition software that determines dimensions, materials, shapes,
colors, sizes, or other aspects of the product depicted in the
image.
[0066] At step 224 and based upon an analysis of the image, the
branched tree of the model is traversed to obtain one or more
potential export codes for the products-to-be-shipped. The
potential export codes are situated at one or more leaves of the
tree and each of the potential export codes has an associated
probability. For example, analysis may indicate that image is of
apparel, which branches into pants, shirts, socks, or other, each
of which may branch into men's woman's and children, and these
further branch into fabric types (e.g., cotton, polyester), and
then finally into leaves which include codes and probabilities.
Again, the end result of a traversal of a tree is to arrive at one
or more leaves. It will be appreciated that only one leaf may be
determined, but in some instances more than one leaf may be the end
result. For example, for some item of apparel that is 50% cotton
and 50% polyester, two leaves may be obtained (one for each
material and each having a code) each having an associated code and
probability for that code being correct.
[0067] At step 226 and at the control circuit, when the associated
probability associated with one of the potential export codes is
greater than a predetermined threshold, the one potential export
code is selected. The threshold may be selected by a user or by an
administrator to mention two examples.
[0068] At step 228 and at the control circuit, when probability of
all the potential export codes is less than a threshold, then one
of the potential export codes is selected based upon a
predetermined criteria. For example, the predetermined criteria may
be an identification of the code with the highest probability.
[0069] At step 230 and at the control circuit, the selected export
code is transmitted to an electronic processing device of the
regulatory source via the electronic communication network. The
electronic processing device may be a computer or electronic server
to mention two examples.
[0070] At step 232, the shipping vehicle responsively exits the
jurisdiction with the products only after the selected export code
is determined. For example, products may be loaded on the vehicle
and the vehicle only allowed to leave the shipping center or
warehouse once a code is obtained. In another example, the vehicle
may be proceeding to an exit point of the jurisdiction (e.g., a
port or border crossing) and the vehicle only allowed to unload or
cross the border when the code has been obtained.
[0071] Referring now to FIG. 3, another example of an approach for
determining the export code of products to be exported from a
jurisdiction is described.
[0072] At step 308, classification data 302, item description data
304, and web-crawling information is pre-processed. The
classification data 302 may include known classification codes,
item descriptions, and department categories. In examples, the
classification data can be from a database in or associated with a
retail store. The item description data 304 may include further
information that describes the product. For example, other keywords
describing products may be utilized that is not included in the
classification data. In examples, this can come from various
sources such as product catalog. The web-crawling information may
include information obtained from web-sites.
[0073] At step 308, any of the data 302, 304, or 306 is cleaned. In
aspects, this step obtains only the correct and required data for a
commodity (e.g., size or ingredient data may be the only data
required to classify a product while other data can be ignored). In
these regards, information can be removed (e.g., brand name and
item title). Spelling errors can be corrected. Abbreviations can be
understood. Split words can be attached. Key words can be
identified. Other cleaning steps are possible.
[0074] Video analytics can be used to recognize product images and
use these to obtain product description data based upon matching
product images. The images may include bar codes, ingredients, or
country of origin to mention a few examples.
[0075] At step 310, machine learning pre-processing may occur. In
this step machine learning may be used to synthesize codes. In
these regards, a table of current Schedule B codes may be used to
find possible codes to add to or modify a model. Additional
examples of codes and relations of these codes to products are
produced. For example, a base export code may be assigned to
numerous types of woman's knit apparel (e.g., the same code may
relate to woman's poncho knit, woman's poncho acrylic knit, and so
forth). Synonyms of words can be used to add to the examples of the
code (e.g., pants and trousers may be the same item so the same
base code may be used to cover both pants and trousers).
[0076] To take another example, the model may not have seen,
considered, or have included a certain item (e.g., an organic
potato) But, there are might be 100 examples of non-organic potato
products with the same or different codes. This step uses machine
learning (using inputs, for example, from the web, store databases,
and other on-line sources) to determine what a code should
(reasonably or potentially) be for a product, e.g., an organic
potato. Put another way, not all products have associated HTS codes
and, in aspects, this step synthesizes code examples for these
products (e.g., code 1234567890 is for an organic potato, and
potentially for other potato products).
[0077] At step 312, the model is trained. In these regards, the
model is modified according to the classification data 302, the
item description data 304, web-crawling information 306, and any
synthesized codes obtained at step 310. In aspects, the
modifications may alter the model by adding, subtracting, or
changing information in the model. When the model is a tree data
structure, the structure (e.g., paths, branches, or leaves) of the
tree may be modified. In examples, new codes may be added to the
tree, the branching of the tree may be modified, or the
probabilities contained at the leaves (and/or the codes at the
leaves) may be changed. Other examples of changes can also be
made.
[0078] When a layered model is used, images pulled from various
sources (e.g., an online catalog) are used to retrain the last
layers of the model in a process known as transfer learning to
detect the edges and shapes in the images.
[0079] At step 314, a new item is to be classified and the model is
tested against the description. In examples, an image can be taken
of the item and/or other information can be entered concerning the
item. One or more codes and their probabilities for being accurate
are obtained by traversing the tree.
[0080] At step 316, it is determined if the probabilities are above
a threshold. If the answer is affirmative, the classification of
the product to a code is accepted at step 318. If the answer at
step 316 is negative, a classification or potential code may be
suggested. This suggestion may be made automatically or by a
human.
[0081] Referring now to FIG. 4, one example of an approach for
pre-processing data is described. At step 402, data is obtained
from an item database 420. For example, the information may include
product descriptions (and other types of information related to a
product) from the database of a retail store. The descriptions, in
examples, may include an item number (e.g., store item number),
item description, department category description, and other
keywords related to the product.
[0082] At step 404, extra information concerning the product is
fetched from the web 422. This information can include information
that is web-scraped from web sites or online catalogs to mention
two examples.
[0083] At step 406, the descriptions (and possibly other
information) obtained from steps 402 and 404 are cleaned.
Translation (e.g., conversion from one language or format into
another language of format) may occur at step 422 and a spell check
may be used at step 426 to mention two examples.
[0084] At step 408, more words are included in the descriptions.
For example, synonyms 428 and other words 430 may be found to
supplement the information obtained in previous steps. In one
example, if "pants" is included in the description the word
"trousers" may also be included.
[0085] At step 410, all of the information may be written to a
pre-process table that is stored in a database 432. This example
continues with the execution of the approach described in FIG.
5.
[0086] Referring now to FIG. 5, one example of an approach for
training a model is described. At step 502, a database 520 (e.g.,
the database 432) is accessed to obtain the classified data for
training the model. The classified data may include the item
descriptions of items and other information.
[0087] At step 504, new examples of export code-item relationships
are synthesized. To take one example, the model may not have seen,
considered, or included a certain item (e.g., export codes a
particular type of woman's shirt). But, there are might be 100
examples of woman's shirt products. This step uses machine learning
(using inputs, for example, from the web, store databases, and
other on-line sources) to determine what a code could potentially
be for the particular type of woman's shirt. In these regards, this
step synthesizes new code examples (e.g., code 1234567890 is for
this particular type of woman's shirt).
[0088] At step 506, the model is trained producing a modified model
508. For example, new leaves may be added to the model if the model
is a tree. In other examples, the branching in the model may be
altered. The classified data obtained from step 502 and the
synthesized examples obtained from step 504 may be used to alter
the model. In one example, the model may not include this
particular product so a new leaf is added at the appropriate place
in the model. Execution can then continue with the approach
described with respect to FIG. 6.
[0089] Referring now to FIG. 6, one example of a classification
approach is described. At step 602, new data (for a new product or
item to be classified) is received from a database 608.
Alternatively, the information may be entered manually. For
example, a user at a portable electronic device may request a code.
The request, for example, may include a photo of the item to be
classified. An analysis of the
[0090] At step 604, the model is used to classify the item. To take
one example, the model is a tree the tree is traversed until one or
more leaves are reached. For example, the item may be a particular
type of woman's shirt (e.g., a woman's all cotton. V-neck, t-shirt)
and the root of the tree may be for apparel. The tree is traversed
until the leaf for this type of woman's shirt is reached. The leaf
will include a potential code and a probability of the code being
accurate for the type of woman's shirt.
[0091] At step 606, the potential code and probability information
are written to a machine learning (ML) classification table in a ML
classification database 610. Execution may then continue with the
approach of FIG. 7.
[0092] Referring now to FIG. 7, one example of an approach for
completing the classification of a product is described.
[0093] At step 702, classified data is fetched from the ML
classification database 701 (e.g., also the ML classification
database 610). The data includes one or more potential codes and
the probabilities associated with the codes. The probabilities may
be percentages obtained from customer feedback or other historical
data that the codes chosen were correct for a particular
product.
[0094] At step 704, it is determined if the probability for the
code is greater than a threshold. If the answer is affirmative, the
item classification (assignment of an export code) is complete at
step 706. This is added to the classification database at step 708.
The thresholds may be dynamic and changeable, and set by a system
administrator.
[0095] If the answer at step 704 is negative, at step 710 a
suggestion is made for the top classification for the product. In
one example and when one more than one code is possible, then the
code with the highest probability may be suggested automatically.
In another example, a human may enter or make a suggestion as to
which of multiple codes may be selected. At step 712, the
classification may be chosen based upon the suggestion. The choice
may be generated automatically or may be made by a human. Execution
then continues with step 708 as has been described above.
[0096] Referring now to FIG. 8, one example of a model is
described. It will be appreciated that the example of FIG. 8 is
only one example of a model, in this case, a tree with a root node,
various branches, and leaf nodes. Other examples of models include
convolutional neural networks, which are known as a class of deep,
feed-forward artificial neural networks utilized for analyzing
visual imagery. It will be appreciated that the present approaches
can be implemented with various types of model data structures or
combinations of data structures.
[0097] In the example of FIG. 8, a root node 802 is for "apparel."
The apparel root node 802 branches into a node 804 (for shirts) and
a node 806 (for pants). The node 804 (shirts), in turn, branches
into a node 808 (t-shirts) and a node 810 (dress shirts).
[0098] The node 808 (t-shirts), in turn, branches to a node 812
(men's t-shirts), a node 814 (woman's t-shirts) and a node 816
(children's shirts). The node 812 branches to node 818 (cotton).
The node 818 branches to a leaf node 822 that has the code (CODE1)
and probability (PROB1) for men's cotton t-shirts. Although not
shown, the leaf nodes may include other information concerning the
product.
[0099] The node 816 (children's t-shirts) branches to a node 820
(50% cotton and 50% polyester) that branches into two leaf nodes
824 (for cotton) and 826 (for polyester). Node 824 includes a CODE2
and PROB2, while Node 826 includes a CODE3 and PROB3.
[0100] It can be seen in this example, that potentially new leaf
nodes can be added (e.g., node 812 may need a polyester-based leaf
with associated code). Other examples of modifications can also be
made to the tree of FIG. 8.
[0101] Also, it will be understood that, in aspects, when an item
is to be classified (tested) then information from then information
from the item (e.g., images, other information) may be obtained an
used to traverse the tree. For example, image information for an
apparel item may be obtained and the image analyzed (using
well-known image analysis techniques) to obtain information
concerning the item to be classified. For example, the analysis may
include first information indicating the item is a shirt, second
information that the item is a t-shirt, third information that the
item is an item of men's clothing, and fourth information that the
item is manufactured from cotton. Once the information is obtain,
the model is accessed and a traversal of the tree made from node
802, to node 804, to node 808, to node 812, to node 818, and
finally to node 822.
[0102] It will also be understood that the example of FIG. 8 does
not show all potential branches or leaves of the tree.
Additionally, it will be appreciated that the tree of FIG. 8 is for
particular types of apparel and that trees for other types of
apparel or other items not related to apparel are possible and the
particular structure of these trees may vary according to various
parameters.
[0103] In some embodiments, one or more of the exemplary
embodiments include one or more localized IoT devices and
controllers (e.g., utilized to obtain data concerning the
products). As a result, in an exemplary embodiment, the localized
IoT devices and controllers can perform most, if not all, of the
computational load and associated monitoring and then later
asynchronous uploading of data can be performed by a designated one
of the IoT devices to a remote server. In this manner, the
computational effort of the overall system may be reduced
significantly. For example, whenever localized monitoring allows
remote transmission, secondary utilization of controllers keeps
securing data for other IoT devices and permits periodic
asynchronous uploading of the summary data to the remote server. In
addition, in an exemplary embodiment, the periodic asynchronous
uploading of data may include a key kernel index summary of the
data as created under nominal conditions. In an exemplary
embodiment, the kernel encodes relatively recently acquired
intermittent data ("KRI"). As a result, in an exemplary embodiment,
KRI includes a continuously utilized near term source of data, but
KRI may be discarded depending upon the degree to which such KRI
has any value based on local processing and evaluation of such KRI.
In an exemplary embodiment, KRI may not even be utilized in any
form if it is determined that KRI is transient and may be
considered as signal noise. Furthermore, in an exemplary
embodiment, the kernel rejects generic data ("KRG") by filtering
incoming raw data using a stochastic filter that provides a
predictive model of one or more future states of the system and can
thereby filter out data that is not consistent with the modeled
future states which may, for example, reflect generic background
data. In an exemplary embodiment, KRG incrementally sequences all
future undefined cached kernals of data in order to filter out data
that may reflect generic background data. In an exemplary
embodiment, KRG incrementally sequences all future undefined cached
kernals having encoded asynchronous data in order to filter out
data that may reflect generic background data. In a further
exemplary embodiment, the kernel will filter out noisy data
("KRN"). In an exemplary embodiment, KRN, like KRI, includes
substantially a continuously utilized near term source of data, but
KRN may be retained in order to provide a predictive model of noisy
data. In an exemplary embodiment, KRN and KRI, also incrementally
sequences all future undefined cached kernels having encoded
asynchronous data in order to filter out data that may reflect
generic background data.
[0104] Those skilled in the art will recognize that a wide variety
of modifications, alterations, and combinations can be made with
respect to the above described embodiments without departing from
the scope of the invention, and that such modifications,
alterations, and combinations are to be viewed as being within the
ambit of the inventive concept.
* * * * *