Note: Descriptions are shown in the official language in which they were submitted.
ARTIFICIAL INTELLIGENCE (Al) BASED DOCUMENT PROCESSOR
[0001]
BACKGROUND
[0002] The evolution of Artificial Intelligence (Al) and machine
learning (ML)
technologies is enabling machines to take over many manual processes. Many
organizations are taking significant strides in this direction by adopting
cognitive and
ML technologies for automating different processes. Machines, such as
computers,
possess different skills than human employees in that the machines are good in
1
Date Recue/Date Received 2022-11-25
terms of precision and consistency. However, machines tend to underperform
employees at tasks that require contextual understanding and complex
communication. Therefore, moving a battery of repetitive tasks to be handled
by
machines provides an advantage in improving the efficiency of repetitive
tasks;
however, these machines often perform poorly when applied to complex task
and/or
tasks requiring contextual understanding.
[0003]
Numerous processes within organizations are driven by documents which
not only serve as the inputs for these processes but are used to collate the
outputs
of the processes. The automation of various tasks can therefore be based
primarily
.. on the processing of the documents involved in the tasks. Many legacy
systems that
were based on paper documents are being digitized and moved online to enable
the
process automation. Forms or documents of various types are widely used for
these
purposes. The documents can include processor-readable documents including
those with structured and unstructured data as well as scanned images,
photographs, etc., which need to be further processed by the machines prior to
collecting and analyzing their data to drive the process automation.
SUMMARY
[0003a]
An example document processing system comprises: at least one data
storage device storing at least a plurality of Machine Learning (ML) models,
and one
or more processors executing machine readable instructions stored in the at
least
one storage device to: receive a request comprising information related to an
2
Date Recue/Date Received 2022-11-25
automated document processing task to be executed, where the automated
document processing task produces an output responsive to the request based at
least on the information provided in the request; extract a process identifier
that
identifies the automated document processing task to be executed from a
plurality
of automated document processing tasks, where the process identifier is
extracted
by preprocessing the request; identify the automated document processing task
associated with the request using the process identifier; retrieve guidelines
associated with the automated document processing task using the process
identifier, where the guidelines include requirements for completing the
automated
document processing task; select a subset of machine learning (ML) models from
the plurality of ML models stored on the at least one data storage device,
where each
ML model of the subset of ML models is trained to extract data for the
requirements
of a corresponding guideline from one or more of the information and at least
one
external data source, wherein a first ML model of the subset of ML models is
trained
to extract data using text and a second ML model of the subset of ML models is
trained to extract data using images; extract data responsive to the
requirements in
the guidelines using the selected subset of ML models from a plurality of ML
models;
determine if a threshold condition for fulfilling the request is met based at
least on
the responsive data extracted by the subset of ML models, where the threshold
condition includes at least a minimum number of the requirements to be met by
the
responsive data; and generate the output responsive to the request based on
the
responsive data extracted by the subset of ML models meeting the threshold
2a
Date Recue/Date Received 2022-11-25
condition, where the output includes one or more of: a recommendation to
approve
the request and a first automatically generated letter, or a recommendation to
reject
the request and a second automatically generated letter.
[0003131 An example machine-implemented method of executing an
automatic
document processing task, comprises: receiving a request comprising
information
related to an automated document processing task, where the automated document
processing task relates to processing an appeal, the appeal associated with a
provider denial of an insurance claim, and the automated document processing
task
producing an output, and the output responsive to the request based at least
on the
information provided in the request; identifying that the automated document
processing task pertains to the appeal of the provider denial of the insurance
claim
using a process identifier, where the process identifier is extracted by
preprocessing
the request; retrieving guidelines associated with the appeal using the
process
identifier, where the guidelines include requirements for completing the
processing
of the appeal; extracting data responsive to the requirements in the
guidelines using
a plurality of machine learning (ML) models, where each ML model of the
plurality of
ML models is trained to extract the responsive data for the requirements of a
corresponding guideline from one or more of the information and at least one
external
data source, wherein a first one of the pluraliry of ML models is trained to
extract
data using text and a second one of the plurality of ML models is trained to
extract
data using images, and each ML model is trained for the data extraction based
at
least on a type of data to be extracted from one or more of the information
and at
2b
Date Recue/Date Received 2022-11-25
least one external data source in response to the requirements; determining
that an
approval threshold for processing the appeal is met based at least on the
responsive
data extracted by the plurality of ML models, the approval threshold includes
at least
a minimum approval score to be met by the responsive data; and generating an
.. appeal letter to the provider, where the appeal letter includes at least a
subset of the
responsive data inserted into a template.
[0003c] An example non-transitory storage medium comprsises machine-
readable instructions that cause at least one processor to: receive a request
comprising information related to an automated document processing task to be
executed, where the automated document processing task produces an output
responsive to the request based at least on the information provided in the
request;
extract a process identifier that identifies the automated document processing
task
to be executed from a plurality of automated document processing tasks, where
the
process identifier is extracted by preprocessing the request; identify the
automated
document processing task associated with the request using the process
identifier;
retrieve guidelines associated with the automated document processing task
using
the process identifier, where the guidelines include requirements for
completing the
automated document processing task; select a subset of machine learning (ML)
models from a plurality of ML models, where each ML model of the subset of ML
models is trained to extract data for the requirements of a corresponding
guideline
from one or more of the information and at least one external data source ,
wherein
a first ML model of the subset of ML models is trained to extract data using
text and
2c
Date Recue/Date Received 2022-11-25
a second ML model of the subset of ML models is trained to extract data using
images; extract data responsive to the requirements in the guidelines using
the
selected subset of ML models from the plurality of ML models; determine if a
threshold condition for fulfilling the request is met based at least on the
responsive
data extracted by the subset of ML models, where the threshold condition
includes
at least a minimum number of the requirements to be met by the responsive
data;
and generate the output responsive to the request based on the responsive data
extracted by the subset of ML models meeting the threshold condition, where
the
output includes one or more of a recommendation to approve or reject a request
and
an automatically generated letter.
BRIEF DESCRIPTION OF DRAVVINGS
[0004] Features of the present disclosure are illustrated by way of
examples
shown in the following figures. In the following figures, like numerals
indicate like
elements, in which:
2d
Date Recue/Date Received 2022-11-25
[0005] Figure 1 shows a block diagram of an Al-based document
processing
system in accordance with an example.
[0006] Figure 2 shows a detailed block diagram of a request
preprocessor in
accordance with the examples disclosed herein.
[0007] Figure 3 shows a detailed block diagram of a data extractor in
accordance
with the examples disclosed herein.
[0008] Figure 4 shows a block diagram of an output generator in
accordance
with the examples disclosed herein.
[0009] Figure 5 shows a flowchart that details a method of executing
the
automatic document processing task in accordance with examples disclosed
herein.
[0010] Figure 6 shows a flowchart that details a method of extracting
the
responsive data using the plurality of ML models in accordance with the
examples
disclosed herein.
[0011] Figure 7 shows a flowchart that details a method of training the
plurality
of ML models for extracting the data in accordance with the examples disclosed
herein.
[0012] Figure 8 shows a flowchart that details a method of managing an
inventory in accordance with the examples disclosed herein.
[0013] Figure 9 shows an example graphical user interface (GUI)
associated
with a provider denial that is generated by the document processing system in
accordance with the examples disclosed herein.
3
Date Recue/Date Received 2021-05-11
[0014] Figure 10 shows a provider denial appeal letter that is
automatically
generated in accordance with the examples disclosed herein.
[0015] Figure 11 illustrates a computer system that may be used to
implement
the document processing system in accordance with examples described herein.
DETAILED DESCRIPTION
[0016] For simplicity and illustrative purposes, the present disclosure
is
described by referring mainly to examples thereof. In the following
description,
numerous specific details are set forth to provide a thorough understanding of
the
present disclosure. It will be readily apparent however that the present
disclosure
may be practiced without limitation to these specific details. In other
instances, some
methods and structures have not been described in detail so as not to
unnecessarily
obscure the present disclosure. Throughout the present disclosure, the terms
"a"
and "an" are intended to denote at least one of a particular element. As used
herein,
the term "includes" means includes but not limited to, the term "including"
means
including but not limited to. The term "based on" means based at least in part
on.
[0017] According to one or more examples described herein, an
artificial
intelligence (Al) based document processing system is described for the
execution
of an automatic document processing task based at least on the information
conveyed in a request for the execution of the automatic document processing
task.
The request can include textual, voice or other data communication providing
some
identifying indicia pertaining to and seeking an output or a result of the
execution of
4
Date Recue/Date Received 2021-05-11
the automatic document processing task. If the request is received as voice
data,
then speech to text application programming interfaces (APIs) can be used to
obtain
the request in a textual format. The request may additionally include one or
more
supportive documents. The request is preprocessed by parsing, tokenizing and
generating parts of speech (POS) data for the tokens. The tokens and the POS
data
are used to identify a specific automatic document processing task to be
executed
from a plurality of automatic document processing tasks that the document
processing system may be configured for. In one example, the automatic
document
processing task can be identified based on a process identifier that can be
determined from the tokens generated from the request. In one example, the
data
generated by preprocessing the request can be used to identify one or more
external
data sources which can provide the process identifier.
[0018] Upon identifying the specific automatic document processing task
to be
executed, guidelines for the execution are retrieved from one or more external
data
sources. The guidelines can include requirements such as data requirements for
the
execution of the automatic document processing task. A plurality of machine
learning (ML) models are used to extract data responsive to the requirements.
Each
of the ML models corresponds to a respective guideline and is trained to
extract data
that fulfill requirements the guideline. Different ML models based on
different
algorithms can be trained to extract the responsive data. The ML model that
corresponds to a guideline will depend on the type of data that is responsive
to that
guideline. In an example, a plurality of ML models can be trained on labeled
training
5
Date Recue/Date Received 2021-05-11
data generated by subject matter experts for each of the plurality of ML
models. In
an example, the labeled training data from different documents in historical
records
includes data that is identified as responsive to each of the requirements of
a given
guideline.
[0019] Responsive data extracted by the plurality of ML models is then
analyzed
for determining if it meets a threshold condition that in turn determines an
output of
the automatic document processing task. In an example, the threshold condition
can
pertain to a minimum number of guidelines or requirements to be met by the
responsive data. However, the guidelines/requirements may be weighted. In such
instances, an approval score can be calculated for the responsive data, for
example,
by aggregating weighted scores of each requirement met by the responsive data.
If
a minimum approval score is achieved by the responsive data, the automatic
document processing task is executed to generate a first type of output. If
the
responsive data fails to meet the requirements and the request does not
achieve the
minimum approval score, then the automatic document processing task is
executed
to generate a second type of output. The outputs thus generated can include a
recommendation to approve or reject the request in one example. In one
example,
an output including an automatically generated letter including the approval
or
rejection decision may also be produced by the document processing system.
[0020] The Al-based automatic document processing system disclosed herein
provides for a technical improvement by enabling more accurate data
extraction, as
compared to conventional techniques, thereby providing better process
automation.
6
Date Recue/Date Received 2021-05-11
Many process automation systems receive certain data inputs, analyze the
received
data and produce certain outputs or automatically execute certain tasks based
on
the analyses of the received inputs. The automatically executed tasks can
include,
but are not limited to, generating recommendations or automatically sending
out
.. certain notifications or communications to preconfigured parties, etc. In
the AI-based
document processing system disclosed herein, the automatically executed tasks
also include automatically generating letters, such as, appeal letters for
provider
denials. As the output that is generated depends on the data inputs provided,
greater
accuracy of the data inputs ensures more accurate outputs. However, the
information can be input to these automation systems in various forms
including
images, documents, databases, voice files, video files, etc. Extracting data
accurately from data sources having a plurality of formats to meet the
requirements
in the guidelines for complex processes such as claims processing, inventory
management, etc., can be a challenge. By employing the plurality of ML models
disclosed herein ensures that accurate data is extracted for that guideline.
For
example, each ML model may be selected and trained to meet one or more
requirements of each of the guidelines. In fact, an ML model can be selected
for
training based on a type of data to be extracted to meet the guideline
requirements.
The document processing system therefore ensures accurate extraction of input
data. As a result, the outputs such as recommendations, communications, auto-
generated letters, etc., are based on accurate input information. In some
instances,
such outputs can also be used to drive downstream processes/systems such as
7
Date Recue/Date Received 2021-05-11
Robotic Process Automation (RPA) systems, Enterprise Resource Planning (ERP)
systems, etc. The AI-based document processing system therefore ensures
accurate data extraction which results in efficient process automation
throughout the
various levels of an organization.
[0021] Figure 1 shows a block diagram of an AI-based document processing
system 100 in accordance with an example. The system 100 receives a request
102
pertaining to one of a plurality of automated document processing tasks that
the
system 100 can be configured to execute. The plurality of document processing
tasks can include processing a claim associated with a disability insurance
and/or
casualty insurance policy in some examples. In another example, document
processing tasks such as processing provider denials of insurance claims may
also
be automatically executed by the system 100 as one of the plurality of
automated
document processing tasks. The request 102 can be received by the system 100
via different modalities, including but not limited to, email, messaging
service, a GUI,
a data store, a portal associated with the document processing system 100, a
social
network platform, etc. The request 102 can include a message 104 with certain
content and may optionally include one or more documents 106 associated with
the
information conveyed in the message 104. If the request 102 is received in a
written
format then the textual content of the message 104 can be extracted directly.
However, if the request 102 is received in a voice/video format, then the
textual
content from the message 104 can be extracted using voice-to-text application
programming interfaces (APIs). The message 104 and the documents 106 can
8
Date Recue/Date Received 2021-05-11
include certain textual content of a plurality of information
types/structures. The
textual content in one or more of the message 104 and the documents 106 can be
presented as structured data with well-formatted information structures such
as
tables, lists, numbered lists, indented textual content, or unstructured data
such as
comma separated values (CSV) data, spreadsheets, etc. For example, if the
request
102 pertains to a worker's compensation claim, the message 104 may include
details
regarding the party making the claim, claim identification details such as
claim
number, policy number, dates, etc. The documentation 106 associated with the
claim can include the claimant's work identification, the claimant's medical
records,
letters from the medical providers such as the doctors, etc. Similarly, if the
request
102 pertains to a casualty insurance claim associated with a theft for
example, the
message 104 may include text describing the claim including claim details such
as
the claim number, policy number, claimant name, place associated with the
theft, the
claimants address, etc. The documents 106 can include a police report, a
formal
valuation of the goods stolen, copies of the policy documents, etc.
[0022] The document processing system 100 processes the message 104
and/or
the documents 106 to extract data 108 required for the execution of the
automated
document processing task specified by the request 102. If the automated
document
processing task pertains to processing of a workers' compensation or casualty
insurance claim, the document processing system 100 can analyze the
information
from the request 102 and one or more external data sources 150 to generate a
recommendation 140 on whether or not the claim can be approved. The external
9
Date Recue/Date Received 2021-05-11
data sources 150 can include information regarding the various policies in
implementation, the policy holders, the requirements associated with the
policies,
and the historical transaction data of the various policy holders, etc. The
external
data sources 150 can include data sources with structured or unstructured data
and
which include information pertaining to specific policies. For example, if the
automatic document processing task pertains to resolving a workers'
compensation
claim, then the external data source that is accessed by the data processing
system
100 can pertain to databases including information regarding workers'
compensation
policies and subscribers of such policies. Similarly, if the automatic data
processing
task pertains to a casualty insurance policy then the external data source
selected
by the data processing system 100 can include information relating to casualty
insurance policies, the guidelines associated with the policies, the
subscribers of
such policies, etc. Therefore, one or more of the external data sources 150
can be
selected for information extraction based on the policy details obtained from
the
request 102.
[0023] Various components of the data processing system 100 can access
or
generate one or more graphical user interfaces (GUIs) 160 which can be used
for
various user interactions. For example, one of the GUIs 160 can be used to
transmit
the request 102 while another one of the GUIs displays the data 108 extracted
from
the request 102. An output 114 that is generated may depend on the automated
data processing task executed by the document processing system 100. If the
automated document processing task 112 relates to an insurance claim, then the
Date Recue/Date Received 2021-05-11
output 114 can include the recommendation 140. If the automated document
processing task relates to a provider denial of a claim, the output 114 may
additionally include an automatically generated letter 116 which appeals the
denial
to the provider along with the requisite documentation. In an example, the
documentation accompanying the letter 116 may include documents extracted from
the request 102 or documents obtained from the external data sources 150.
Output
114 can include other types of data and/or information based on a given
configuration of system 100.
[0024] The document processing system 100 includes a request
preprocessor
122, a process analyzer 124, and an output generator 142. The request
preprocessor 122 processes the request 102 to obtain the data 108 included in
the
request 102. In an example, the request preprocessor 122 can employ techniques
such as, but not limited to, parsing, tokenizing and parts of speech (POS)
tagging on
the text included within the message 104 and/or the documents 106. In an
example,
the document processing system 100 can be coupled to a data store 170 for the
storage of information that is generated and used by the document processing
system 100 during the execution of the various automatic document processing
tasks. Accordingly, the data 108 obtained by the request preprocessor 122 can
be
stored within the data store 170.
[0025] The process analyzer 124 accesses the data 108 obtained by the
request
preprocessor 122 to identify an automatic document processing task to be
executed.
As mentioned above, the data 108 can include a process identifier 132 relating
to
11
Date Recue/Date Received 2021-05-11
the processes to be executed. Depending on the automatic document processing
task to be executed one or more of the process identifier 132, e.g., certain
keywords,
member identifiers, etc. While the description herein generally refers to the
process
identifier 132 as enabling identification of the automatic document processing
task,
other process identifiers may also be used in accordance with some examples
disclosed herein. In an example, a policy can pertain to an insurance policy
associated with a workers' compensation claim. Upon the process identifier 132
identifying the policy pertaining to the request 102, the guidelines retriever
126
retrieves guidelines 194 associated with the policy. In an example, the
guidelines
194 can be retrieved from one of the external data sources 150 that pertains
to the
policy. Therefore, different policies may necessitate retrieval of the
guidelines 194
from different external data sources. In an example, the guidelines 194
retrieved
from one of the external data sources 150 may be cached temporarily on the
data
store 170 during the execution of the automatic document processing task 112.
The
guidelines 194 can include certain data requirements that need to be met if
the
automatic document processing task is to be executed. Referring again to the
worker's compensation request example, the corresponding guidelines can
include
data requirements for the claimant's information such as name, social security
number, address, employer information, type of job, date of injury, nature of
injury,
etc. In addition, the guidelines 194 can also include requirements for
clinical data
and medical history of the claimant. The responsive data 196 per the
requirements
of the guidelines 194 is extracted from one or more of the data 108 and the
external
12
Date Recue/Date Received 2021-05-11
data sources 150 by the data extractor 128 using a plurality of ML models 138.
In
an example, each of the requirements and/or the guidelines 194 can be
associated
with a corresponding ML model that is trained to identify information
responsive to
the requirement. For example, if a guideline includes multiple requirements,
then
respective multiple ML models are used for extracting the data responsive to
that
guideline. If the guideline includes only one requirement, then a single ML
model
may be used for the extraction of responsive data for that guideline. The
responsive
information can include multiple pieces of data that is gathered by the
corresponding
ML model from the different data sources. The responsive information can be
presented via one of the GUIs 160 for validation.
[0026] In an example, the output generator 142 can be configured to
present the
results from the data extractor 128 for validation. The output generator 142
can be
further configured to generate a recommendation to approve or reject the
request
102 based on the responsive data 196 and letters may be automatically
generated
to convey the output 114. If the request 102 pertains to a workers'
compensation
claim or a casualty insurance claim, the output generator 142 can be
configured to
determine whether the request 102 satisfies certain threshold criteria. Based
on the
request 102 satisfying the threshold criteria, the recommendation 140 to
approve or
reject the request can be generated. Certain automatic actions can be executed
by
the document processing system 100 upon validation of the responsive data 196.
The automatic actions can be executed based on the type of document processing
task specified in the request 102. In an example, a recommendation to approve
the
13
Date Recue/Date Received 2021-05-11
request 102 can cause the document processing system 100 to produce an
automatically generated letter 116 to include an approval of the request,
while a
recommendation to reject the request 102 can cause the document processing
system 100 to produce the automatically generated letter 116 with a rejection
of the
request 102. Moreover, when the automatic document processing task 112
pertains
to processing a provider denial, another one of the automatic actions that the
output
generator 142 can be configured to execute includes producing an automatically
generated letter 116 for appealing the provider denial.
[0027] The document processing system 100 further includes a model
trainer
144 for training the plurality of ML models 138. The model trainer 144 can
employ
training data 146 to train the plurality of ML models 138 to extract the
responsive
data 196 for the guidelines 194. The plurality of ML models 138 are trained
via
supervised training methods in one example. The training data 146 for the
supervised training can be generated for each requirement within the
guidelines 194
by identifying from different data sources various pieces of information that
are
responsive to that requirement. In fact, it can happen that the same
information can
be conveyed in different formats. For example, proof of an injury can be
provided as
text describing the injury or as an image of the injury. Accordingly, multiple
ML
models of the plurality of ML models 138 can be trained to identify responsive
data
for the same requirement. The contributions of the multiple ML models for that
requirement can be considered in terms of weightage that the request 102 is
assigned under that requirement when processing for approval/rejection
thresholds.
14
Date Recue/Date Received 2021-05-11
[0028] The ML model to identify textual information can be trained on
identifying
different words for the same condition, using contextual data to identity the
condition,
etc. In an example, the ML model to identify images pertaining to the
condition can
be trained on different images of that condition taken from different people
and
different angles accompanied with an instruction that the images pertain to
the
specific condition. Similarly, an ML model can be trained on multiple data
instances
of a particular data type that can occur within the training data 146. As
newer
guidelines and data requirements are added/updated, or newer insurance
products
are introduced, new ML models can be trained or existing ML models can be
upgraded in accordance with the methods outlined herein to extract the data
responsive to the newer guidelines.
[0029] Figure 2 shows a detailed block diagram of the request
preprocessor 122.
The request preprocessor 122 can include a document extractor 202, a parser
204,
a tokenizer 206, and a POS tagger 208. The document extractor 202 extracts
documents associated with the request 102. Different documents can be
associated
with the request 102 based on the automatic document processing task 112. The
documents 106 can be transmitted as attachments when the request 102 is
received
in an email or as accompanying messages, e.g., when the request 102 is
received
via a chat window or a GUI or a portal for uploading the documents 106. The
parser
204 parses the text included in one or more of the message 104 and the
documents
106 of the request 102. The tokenizer 206 can produce word tokens from the
output
of the parser 204. Tokens can be further processed to remove stop words,
Date Recue/Date Received 2021-05-11
punctuations, etc. The POS tagger 208 tags each of the tokens with the POS
information. Different policies pertaining to different automatic document
processing
tasks, e.g., workers' compensation, may have identifiers of string types that
are
different from the identifiers of casualty insurance. Therefore, a policy
identifier
(which serves as the process identifier 132) having a specific string type can
be
identified via pattern matching techniques based on the tokens, and an
automatic
document processing task to be executed can be identified from the specific
policy
identifier. The tokens from the request preprocessor 122 along with the POS
information enable obtaining the process identifier 132 and other information
such
.. as process keywords that allow the process analyzer 124 to identify the
automatic
document processing task 112 to be executed. The guidelines retriever 126 can
select one of the external data sources 150 that correspond to the automatic
document processing tasks 112 to obtain details pertaining to the process
identifier
132 such as the associated guidelines.
[0030] Figure 3 shows a detailed block diagram of the data extractor 128.
The
data extractor 128 includes a model selector 302 and a response retriever 304.
The
guidelines 194 that are retrieved include various requirements that are to be
met in
order to process the request 102. The requirements can include data
requirements
for identifying information of the claimant, medical details if the automatic
document
processing task 112 pertains to a workers' compensation claim or health
related
claim, the provider data, the dates associated with the claim, the employer
information, etc. If the process identifier 132 pertains to a casualty
insurance claim,
16
Date Recue/Date Received 2021-05-11
e.g., property theft, then the guidelines 194 can have discrete data
requirements for
details of the claimant such as name, address, social security number,
information
about stolen item(s), date the theft occurred, location at which the theft
occurred,
date of purchase of the stolen items, a complaint number of a police report
pertaining
to the stolen items, images of the stolen items, color or other identifying
indicia or
attributes of the stolen items, etc. As mentioned above, certain data
requirements
can have multiple responsive data items. Each discrete piece of data
responsive to
the data requirement can have a corresponding ML model of the plurality of ML
models 138 trained to identify that discrete piece of data. The plurality of
ML models
138 can include classification models such as support vector machines (SVMs),
random forests, linear classification models such as K-means, logistic
regression
models, etc. Based on the determined type of data responsive to a specific
data
requirement, ML models such as convolutional neural networks (CNNs), recurrent
neural networks (RNNs), Long Short Term Memory (LSTM) or even ensemble
models can be trained to identify the responsive data 196. For example, CNN
based
models can be trained to identify images while LSTM which is a special
category of
RNN can be employed to understand the context within a whole
paragraph/sentence
to determine whether a condition needs to be presented to a coder for
associated
with a specific condition code. A logistic regression model can be trained for
extracting data pertaining to a categorical variable where the categorical
variable
constitutes at least a part of the responsive data 196 for one of the
guidelines.
17
Date Recue/Date Received 2021-05-11
[0031] The model selector 302 can be configured to select a subset of
one or
more ML models 352 from the plurality of ML models 138 for obtaining the
responsive
data 196 for the requirements specified in the guidelines 194. In an example,
the
model selector 302 can be configured to select the correspondingly trained ML
model
for a given discrete piece of data. For example, if the data requirement
pertains to a
social security number, a specific ML model trained to identify the social
security
number from one or more of the request 102 or the associated external data
sources
150 is selected by the model selector 302. Similarly, if an image data
including an
X-ray of a specific broken bone is the responsive data to be identified for a
requirement, then a ML model such as a CNN trained to identify images of that
specific broken bone from one or more of the request and the external data
sources
150 can be selected by the model selector 302.
[0032] The response retriever 304 employs the subset of ML models 352 on
one
or more of the request 102 and the external data sources 150 to extract the
responsive data 196. In an example, the request 102 with or without the
documents
106 can include all the responsive data 196. In an example, the request 102
may
be a follow up communication continuing the correspondence regarding the
matter
associated with the automatic document processing task 112. For example, the
request 102 can be a reminder regarding a workers' compensation claim. The
request 102 may therefore include minimum process-identification information,
such
as, the claim number. In such instances, the claim number or other identifying
indicia
extracted from the request 102 can be used to obtain additional information
from one
18
Date Recue/Date Received 2021-05-11
or more of the external data sources 150, which may store the other required
information pertaining to the request 102 to generate the responsive data 196.
It
may be appreciated that there may be specific data sources from the external
data
sources 150 corresponding to the request 102. For example, when the claim
number
corresponds to a workers' compensation claim, only the data sources
corresponding
to the workers compensation products are processed by the one or more ML
models
352. Metadata associated with the request 102 such as the date/time the
request
102 was received, the modality (i.e., email, fax transmission, etc.) in which
the
request was received, the sender of the request 102, any name(s) included in
the
request 102 (if different from the sender of the request), etc., can also be
used to
determine the responsive data 196. The responsive data thus retrieved are
transmitted to the output generator 142 for presentation in accordance with
the
automatic document processing task 112.
[0033] Figure 4 shows a block diagram of the output generator 142 in
accordance
with the examples disclosed herein. The output generator 142 includes a
recommendation generator 402, a data validator 404 and a letter generator 406.
The
recommendation generator 402 includes a threshold analyzer 422 and a
recommendation provider 424. If the automatic document processing task 112
pertains to settling an insurance claim, the recommendation generator 402 can
be
configured for the generation of a recommendation to approve or reject the
claims
based on the responsive data 196. The threshold analyzer 422 determines if an
approval threshold 452 is satisfied to generate an approval recommendation.
The
19
Date Recue/Date Received 2021-05-11
threshold analyzer 422 can analyze different conditions based on the
guidelines 194
that are satisfied. The approval threshold 452 can include calculating an
approval
score for the responsive data and a minimum approval score to be met for the
claim
to be approved. The approval score can include a weightage to be assigned to
each
of the requirements depending on the responsive data 196. The total weightage
for
the requirements in the guidelines can be designated as the approval score.
The
minimum approval score can be set empirically by human reviewers in an
example.
In another example, a minimum approval score can be programmatically set using
system 100 or another device.
[0034] Reverting to the workers' compensation example, each of data such as
the claimant's name, address, employer, etc., that matches the data in the
records
on the external data sources 150 can be assigned certain points. In addition,
any
medical requirements that are met can also be assigned certain points. For
example,
one of the guidelines 194 can be associated with an administrative requirement
regarding the claimant's time period of employment while another one of the
guidelines can pertain to a medical requirement for a confirmation regarding
the
chronic medical condition of the claimant. The administrative requirement can
carry
less weight as compared to the medical requirement. Even for the same
requirement, e.g., the administrative requirement, positive and negative
responses
can carry different points. The points thus assigned to each of the guidelines
194
can be further multiplied with the corresponding weight of the guideline and
aggregated across the guidelines 194 to obtain the approval score. Based on
the
Date Recue/Date Received 2021-05-11
comparison of the approval score with the minimum approval score, the
threshold
analyzer 422 determines if the approval threshold 452 is satisfied. If yes,
the
recommendation provider 424 generates a recommendation for approval of the
claim, else a recommendation for rejection of the claim can be generated.
[0035] The responsive data 196 obtained by the data extractor 136 along
with
the recommendation 140 can be presented for validation via a validation GUI
454
generated by the data validator 404. In an example, the validation GUI 454 can
present one or more of the discrete data items from the responsive data 196 in
an
editable format so that a human reviewer who is validating can make any
necessary
changes to the data. In an example, the validation GUI 454 can include two
portions
where the extracted data is presented in a first portion and a corresponding
view of
the original data source, such as, a document, a database table or an image,
etc.,
obtained from either the request 102 or the external data source from which
the data
piece was extracted can be displayed in a second portion. In an example, the
validation GUI 454 can also include the recommendation 140 to approve or
reject a
claim associated with the request 102. A human validator may agree or disagree
with the recommendation 140. The feedback from the human validator including
any
edits to the responsive data 196 can be provided to the document processing
system
100 for further training.
[0036] The output generator 142 additionally includes a document generator,
such as letter generator 406, that can be activated for certain automatic
document
processing tasks such as provider denials. When a medical insurance claim,
e.g.,
21
Date Recue/Date Received 2021-05-11
the workers' compensation claim is denied, the automatic document processing
task
112 can pertain to analyzing the denial. If the document processing system 100
generates the recommendation 140 to withdraw the denial upon analyzing the
guidelines 194 and the responsive data 196 as disclosed herein, the automatic
letter
generator 406 can be activated to automatically generate an appeal letter
i.e., the
automatically generated letter 116 appealing the denial. In an example, the
automatically generated letter 116 can be generated by substituting one or
more
data items from the responsive data 196 into a letter template.
[0037] Figure 5 shows a flowchart 500 that details a method of
executing the
automatic document processing task 112 in accordance with examples disclosed
herein. The method begins at 502 with receiving the request 102 for the
execution
of the automatic document processing task 112. The request 102 includes
information such as one or more of the message 104 and the documents 106. The
request 102 is preprocessed at 504 to extract the documents 106 (if any) and
to
obtain the data 108 such as tokens or POS tags. The data 108 thus extracted is
analyzed to obtain the process identifier 132 at 506. The process identifier
132 is
employed to identify a process to be executed. In an example, the data 108 can
include the process identifier 132 which can be employed at 508 to determine
the
automatic document processing task 112 to be executed.
[0038] The guidelines 194 including the requirements for the execution of
the
automatic document processing task 112 are retrieved at 510. In an example,
the
guidelines 194 can include requirements for all policies under a specific
scheme.
22
Date Recue/Date Received 2021-05-11
However, the guidelines 194 can also include data requirements specific to a
policy
represented by the process identifier 132. For example, in instances where
certain
data requirements essential for executing the automatic document processing
task
112 were not provided in earlier communications, policy-specific data
requirements
to the guidelines 194 may be added programmatically or via a human reviewer.
At
512, the responsive data 196 for the guidelines 194 is extracted from one or
more of
the request 102 and the external data sources 150 using at least a subset of
the
plurality of ML models 138. Each of the selected subset of ML models is
trained to
extract data responsive to one of the guidelines 194. At 514, the responsive
data
196 is evaluated to determine if it meets, or satisfies, the approval
threshold 452 for
determining the output of the automatic document processing task 112. If the
automatic document processing task 112 pertains to an insurance claim, the
approval threshold 452 can represent a certain score attained by the claim
which
causes the claim to be eligible for approval. If the responsive data 196 meets
the
approval threshold 452, the automatic document processing task 112, such as
generating a recommendation for approval of the claim associated with the
request
102 is executed at 516. The automatically generated letter 116 can be produced
at
520 to include the claim approval/disapproval. For example, a first
automatically
generated letter including the claim approval or a second automatically
generated
letter including the claim rejection may be produced at 520. Similarly, an
appeal
letter can be automatically generated upon the approval threshold 452 being
met by
the responsive data 196 in the instances such as provider appeals.
23
Date Recue/Date Received 2021-05-11
[0039] If it is determined at 514 that the approval threshold 452 is
not met, then
the output 114 pertaining to a recommendation to reject the claim associated
with
the request 102 may be produced at 518 recommending claim denial. In the case
of provider appeals, the automatically generated letter 116 appealing the
provider
denial is not produced when the responsive data 196 fails to meet the approval
threshold 452.
[0040] Figure 6 shows a flowchart 600 that details a method of
extracting the
responsive data 196 using the subset of ML models 352 in accordance with the
examples disclosed herein. Although the method describes application of ML
models serially for data extraction, it can be appreciated that this is for
illustration
purposes only and that the subset of ML models 352 can be used simultaneously,
e.g., in parallel, for the data extraction. At 602, the plurality of ML models
138
corresponding to each of the guidelines 194 that are trained to identify data
responsive to the requirements of the guidelines 194 are accessed. At 604, one
of
the guidelines 194 is selected for processing. The corresponding ML model(s)
that
are trained to extract the data responsive to the selected guideline are
further
selected at 606. If, for example, the selected guideline includes requirements
for
more than one data item, then more than one ML model can be selected to
extract
data at 606. In an example, the model selector 302 may store a data structure
such
as a table that specifies the ML model(s) to be selected for a given guideline
and the
selection may be carried out in accordance with the information in the table.
Each
time one or more of a requirement and the ML models associated with the
24
Date Recue/Date Received 2021-05-11
requirement are updated, the table can be correspondingly updated. The
selected
ML model(s) are applied to one or more of the request 102 and the external
data
sources 150 at 608 and the data responsive to the requirements is obtained. At
610,
it is determined if more guidelines remain to be processed. If yes, the method
moves
to 604 to select the next guideline for processor. If it is determined at 610
that no
more guidelines remain for processing the method terminates on the end block.
[0041] Figure 7 shows a flowchart 700 that details a method of training
the
plurality of ML models 138 for extracting the data in accordance with the
examples
disclosed herein. At 702, one of the plurality of ML models 138 corresponding
to one
of the guidelines 194 is accessed. Each of the guidelines 194 can have
corresponding one or more of the plurality of ML models 138 trained to provide
data
responsive to the guidelines based on a type of data that is expected. If the
guideline
expects text data in specific patterns such as social security numbers, dates,
policy
numbers, etc. then classification ML models suitable for prediction of textual
data
can be selected and trained to identify textual data in the specific pattern.
If the
guideline requires image data to be identified, then image classification ML
models
such as CNNs, deep learning networks (DLNs), etc. can be employed. In certain
other examples, ensemble models based on two or more ML algorithms may also
be employed. Accordingly, large volumes of training data for each of the
plurality of
ML models 138 that correspond to the type of data to be predicted by the ML
model
needs to be generated. At 704, data that was gathered and/or generated during
prior
document processing tasks which are similar to the document processing task
112
Date Recue/Date Received 2021-05-11
can be accessed. For example, documents pertaining to previously approved,
settled, or rejected insurance claims can be digitized (i.e., scanned and text
made
machine-readable and machine searchable) and used to generate the training
data
146. The training data 146 thus generated can be split into training data and
test
data. The collected data is used to train the plurality of ML models 138 and
the test
data can be used to test the trained ML model. Generally, the collected data
is
partitioned so that 80% of the data is training data while 20% of the data is
used for
testing the trained model.
[0042] The training data is labeled as accurate or inaccurate response
to the
guideline at 706 and provided to train the ML model at 708 for supervised
learning.
The trained ML model is tested with the testing data for accuracy at 710. It
is
determined at 712 if an adequate level of accuracy is achieved. The trained
model
is employed by the document processing system 100 as one of the plurality of
ML
models 138 for data extraction at 714 if adequate accuracy is obtained, else
the ML
model is further trained at 716 and its accuracy is again determined. The
cycle may
be repeated until satisfactory accuracy is achieved for the ML model.
[0043] The document processing system 100 can be used for executing
various
automatic document processing tasks in different organizations such as
insurance
companies, hospitals, pharmacies, etc. One of the automatic tasks that can be
executed by the document processing system 100 in hospitals, doctors' offices,
etc.
includes inventory management.
26
Date Recue/Date Received 2021-05-11
[0044] Figure 8 shows a flowchart 800 that details a method of managing
an
inventory by executing inventory management tasks in accordance with the
examples disclosed herein. The document processing system 100 receives request
102, which can pertain to a query regarding the size of an order for a product
in an
inventory, e.g., syringes or other medical equipment. At 804, the data 108
regarding
the inventory query and the product associated with the query is extracted by
the
request preprocessor 122 from the request 102. The tokens, POS tags and other
output from the request preprocessor 122 is accessed by the process analyzer
124
to determine at 806 that the automatic document processing task 112 pertains
to
obtaining a demand projection for the product specified in the request 102
using, for
example, a product id or a product code. The process analyzer 124 can use
techniques such as but not limited to natural language processing (NLP) for
analyzing the output of the request preprocessor 122 and identifying the
process to
be executed. The guidelines 194 for executing a process to obtain demand
projections for a product are retrieved at 808 using the product id. The
guidelines
194 can include requirements for current inventory levels of the product
identified by
the product id as well as requirements for predictions for prospective demand
for the
product for a predetermined time period based on current requirements.
Accordingly, different programming constructs such as database access scripts,
ML
models, etc. can be used to obtained the responsive data 196 for the
guidelines 194.
The current inventory levels or stock levels can be obtained via running a
query
against the inventory database while a subset of one or more ML models 352
that
27
Date Recue/Date Received 2021-05-11
are trained to predict the prospective demand for the product are selected at
810
from the plurality of ML models 138. The responsive data 196 including the
current
inventory levels of the product and the prospective demand for the product is
obtained at 812. ML models based on approaches such as but not limited to,
time
series, linear regression, feature engineering, and random forests can be
trained to
project the prospective demand for the product using the current requirements.
The
recommendation 140 produced at 814 can include the quantity of the product to
be
ordered to meet the prospective demand in view of the current inventory
levels.
[0045] Figure 9 shows an example GUI 900 which can be a GUI 160
generated
.. by the document processing system 100 for an automatic document processing
task
associated with a provider denial in accordance with the examples disclosed
herein.
The GUI 900 includes certain features that may be commonly implemented across
GUIs 160 generated for the various document processing tasks. These features
can
include a left-hand side (LHS) panel 902 which provides access to different
sections
of information extracted from the various documents that were received and
processed in connection with the provider denial document processing task. For
example, sections can include a claims history 922, a denied details 924, and
a
clinical review 926. As the GUI 900 pertains to the provider denials process,
if the
recommendation 140 suggests that the provider denial is improper or if a human
validator deems the provider denial to be improper, a generate letter button
952 on
the LHS panel 902 can be activated to automatically generate a letter
appealing the
denial. The GUI 900 also includes a right-hand side (RHS) panel 904 that
displays
28
Date Recue/Date Received 2021-05-11
the relevant information based on the selections made in the LHS panel 902. In
an
example, the relevant information can be shown in the RHS panel 904 from the
original documents or original data source from which the relevant information
was
extracted. Also, the RHS panel 904 can highlight different attributes 942 of
an entity
associated with the document processing task. The GUI 900 displays the
attributes
942 such as but not limited to, subscriber ID, last name, first name, Medicare
no.,
phone number, date of birth, etc., of a subscriber associated with the
provider denials
process. A searchable representation of the documents 106 accompanying the
request is therefore generated and displayed on the GUI 900.
[0046] Figure 10 shows an example provider denial appeals letter 1000 that
is
automatically generated in accordance with the examples disclosed herein. The
appeals letter 1000 includes a patient details section 1002 that is
automatically filled
with the attributes 942 gathered from the patient's file or documents. In
addition to
general attributes such as patient name, date of birth, member id, etc.,
specific
details regarding a particular service pertaining to the denied matter such as
the
Hospital, dates of service, billed amount, etc., are also included in the
patient details
section 1002. In an example, a template of the letter may be stored in one of
the
data store 170, or the external data sources 150 can be retrieved. The
template
includes predetermined or standard language appealing the provider denial with
place holders within the standard language for receiving at least a subset of
the
responsive data extracted from one or more of the request 102 and the external
data
sources 150. For example, the patient details section 1002 may include such
place
29
Date Recue/Date Received 2021-05-11
holders which are completed with the corresponding patient details retrieved
from
the request 102 and/or the external data sources 150. In an example, the
tokens
corresponding to the place holders can be identified using named entity
recognition
(NER), tokens from the responsive data 196, and the letter 1000 is generated
with
the tokens inserted or included in the corresponding place holders.
[0047] The body of the letter 1004 includes the details of the service
that was
denied to the patient and the pertinent information identified by the document
processing system 100 where John who is 66 years old was categorized as being
50 years old as one of the reasons for withdrawal of the denial. Again, the
template
may include place holders that can be configured with scripts to receive the
relevant
patient details. When the human reviewer presses the submit button 1006, the
denial
appeals letter 1000 will be submitted to the health plan provider.
[0048] It can be appreciated that although the automatically generated
letter 116
is described herein as a document with data inserted therein, other examples
of
automatically generated letters can include any document, file, etc.,
containing the
relevant information in digital or hardcopy form.
[0049] In addition to the automated document processing tasks described
above,
the document processing system 100 can be employed in different fields for the
automatic execution of various document processing tasks as outlined below.
[0050] Embodiments of the invention can be configured to address health
payer
use cases, such as provider claims and disputes. For example, document
processing system 100 can be configured to review provider disputes and
claims.
Date Recue/Date Received 2021-05-11
For example, provider responses from claim denials can be reviewed for
adjudication. Appeal letters for provider denials or other letters can be
automatically
generated as described above. Another health payer application of the document
processing system 100 can include provider data management. For example, the
.. document processing system 100 can be configured to review documents that
are
required to maintain, terminate, or add new provider data, such as data a
doctor,
nurse, lab technician, etc. The message 104 in the request 102 can include
particular keywords such as but not limited to, "maintain", "terminate" or
"add" new
provider data which may be specified in one or more of the message 104 or the
documents 106. Upon retrieving the guidelines 194 for the particular process
and
extracting the responsive data 196, the corresponding information in the
external
data sources 150 can be updated.
[0051] In some examples, the document processing system 100 can be used
to
address health provider use cases such as Starts and Healthcare Effectiveness
Data
and Information Set (HEDIS) Chart Review. Employers and individuals use HEDIS
to measure the quality of health plans. HEDIS measures how well health plans
give
service and care to their members. In addition to evaluating healthcare plans,
the
document processing system 100 can also be configured to review medical
records
and Health Level Seven (HL7) messages for quality measures. HL7 International
specifies a number of flexible standards, guidelines, and methodologies by
which
various healthcare systems can communicate with each other. Such guidelines or
data standards are a set of rules that allow information to be shared and
processed
31
Date Recue/Date Received 2021-05-11
in a uniform and consistent manner. These data standards are meant to allow
healthcare organizations to easily share clinical information. Again, the
request 102
can include medical records and/or HL7 messages while the quality measures
(i.e.,
the guidelines 194) can be retrieved from the external data sources 150. The
document processing system 100 can extract the responsive data 196 for the
requirements specified in the quality measures and generate the recommendation
140 on whether the medical records or the HL7 messages meet the requirements
of
the quality measures.
[0052] The document processing system 100 also finds application in the
risk
adjustment chart review. For example, the document processing system 100 can
be
configured to review medical records and/or HL7 messages. The guidelines 194
can
include requirements to determine if the risk adjustment reimbursement was
received. Based on the responsive data 196 retrieved by the data extractor
128, the
output 114 can include a recommendation on whether or not the risk adjustment
reimbursement was received.
[0053] Another application of the document processing system 100 in the
health
provider use cases includes utilization management document intake. The
document processing system 100 can be configured to perform indexing of
authorization forms for prior, post, and concurrent review.
[0054] The document processing system 100 can also be employed to
restructure unstructured data into Electronic Medical Records (EMR). EMRs
typically contain general information such as treatment and medical history
about a
32
Date Recue/Date Received 2021-05-11
patient. By implementing EMR, patient data can be tracked over an extended
period
of time by multiple healthcare providers. Unstructured data and documents can
be
restructured into [MR profile using the document processing system 100.
[0055] The document processing system 100 can be configured for clinical
coding/billing to review International Classification of Diseases (ICD) 10
codes (or
ICD 9 codes whichever is applicable) and flag for charging. The ICD-10 codes
are
broken down into chapters and subchapters and include a letter plus two digits
to the
left of the decimal point, then one digit to the right. The new system allows
for a
more specific diagnosis. When a medical service provider submits a bill to
insurance
for reimbursement, each service is described by a common procedural technology
(CPT) code, which is matched to an ICD code. The document processing system
100 can receive the provider's bill in the request 102. The data 108 is
extracted from
the request 102. The guidelines 194 include requirements where the CPT code
from
the bill in the request 102 be aligned with the corresponding ICD code. If the
two
codes don't align correctly with each other, a recommendation can be generated
to
reject the payment. In other words, if the service isn't one that would be
typically
provided for someone with that diagnosis, insurance will not pay. Therefore,
the
document processing system 100 can analyze the codes and correlate the tests
to
diagnoses to ensure correct reimbursement. In case there are any
discrepancies,
the reimbursements may be denied and the provider denial process may be
activated at that point.
33
Date Recue/Date Received 2021-05-11
[0056] The document processing system 100 can be configured to process
social
determinants of health to isolate determinants to improve health outcomes.
[0057] The document processing system 100 can be configured for use in
precision medicine to isolate determinants for better health outcomes and to
provide
.. a tailored treatment to individuals based upon available clinical data down
to the
genomic level.
[0058] Non-clinical applications for the document processing system 100
can
include supply chain management for forecasting usage of medical supplies as
detailed above, insurance credentialing, auto claims processing, mortgage/loan
application processing, insurance data management, etc. For example, the
document processing system 100 can be configured to perform a clinical review
of
authorization forms for prior, post, and concurrent review.
[0059] The document processing system 100 can be employed in the
automotive
sector for reviewing the information provided to support auto insurance
claims. The
request 102 can include information and documents related to an auto insurance
claim. Based on the information conveyed in the request 102, the process
identifier
132 and hence the guidelines 194 are retrieved as disclosed herein. The data
extractor 128 can extract the responsive data 196 using the plurality of
models 138
which can be trained on prior auto insurance claim data as described above.
.. Depending on whether the responsive data 196 meets the approval threshold
452,
the recommendation 140 may suggest approving or rejecting the auto insurance
claim.
34
Date Recue/Date Received 2021-05-11
[0060] The document processing system 100 can be configured to review
documents to support loan or mortgage applications in yet another non-clinical
application. Based on the information conveyed in the request 102 which can
include
the documents 106 that support the loan/mortgage application, the guidelines
194
are retrieved as disclosed herein using one or more of NER and NLP. The data
extractor 128 can extract the responsive data 196 using the plurality of
models 138
which can be trained on prior loan/mortgage data as described above. Depending
on whether the responsive data 196 meets the approval threshold 452, the
recommendation 140 may suggest approving or rejecting the loan/mortgage
application.
[0061] The document processing system 100 can be configured for
management
of structured and unstructured information relating to insurance policies,
applications, and claims in a use case.
[0062] Another use case for the document processing system 100 can
include
identification of potential fraud in insurance claims.
[0063] The document processing system 100 can be configured for
extracting
data from unstructured documents and transforming it into evidence for
decision
making in yet another use case. If the request 102 includes one or more of the
message 104 and the documents 106 in the form of unstructured data, the
request
preprocessor 122 and the data extractor 128 can be configured to extract the
responsive data 196 which enables generating the recommendation 140 regarding
the request 102 based on the responsive data 196 meeting the requirements in
the
Date Recue/Date Received 2021-05-11
guidelines 194 which provide the requirements for the presence or absence of
real-
world evidence.
[0064] The document processing system 100 also finds applications in the
fields
of clinical research and patient matching. The document processing system 100
can
be configured for mining scientific literature and matching appropriate
patients for
clinical trials. If the request 102 includes one or more of the message 104
and the
documents 106 pertaining to patients. An identifier for the clinical trials
can be
extracted to retrieve the guidelines 194 for selecting patients for the
clinical trials.
The responsive data 196 regarding each of the patients can be extracted from
one
.. or more of the external data sources 150 or the information provided with
the request
102 using the plurality of ML models 138 for the requirements outlined in the
guidelines 194. The plurality of ML models 138 can be trained on prior patient
records to identify conditions within the patient records that would make a
patient a
good candidate for the specific clinical trial. Those patients whose data
matches the
requirements as determined by the approval threshold 452 can be recommended
for
the clinical trial by the output generator 142.
[0065] Regulatory Compliance forms yet another use case for the document
processing system 100. Particularly, the document processing system 100 can be
configured to find, highlight, and extract key data within regulatory
documents which
may be received in the request 102. The guidelines 194 can specify the
requirements for regulatory compliance. The data extractor 128 can employ the
plurality of ML models 138 which are trained to extract the responsive data
196 in
36
Date Recue/Date Received 2021-05-11
accordance with the requirements. The key data (i.e., the responsive data 196)
can,
therefore, be extracted from the regulatory documents.
[0066] Other use cases for the document processing system 100 include
precision medicine, drug discovery, and pharmaceutical covigilence.
[0067] In the field of precision medicine, the document processing system
100
can be configured for providing tailored treatment to individuals based upon
available
clinical data down to the genomic level.
[0068] The document processing system 100 can be configured for use in
drug
discovery by employing NLP to extract previously discovered chemical reactions
to
evaluate the need for experiments.
[0069] As an application for use in pharmaceutical covigilence, the
document
processing system 100 can be configured for identifying potential safety
opportunities earlier in the drug development process and achieving faster
Adverse
Drug Reactions (ADR) and Medical Device Reporting (MDR) determinations and
improved safety profiles.
[0070] Other clinical use cases for the document processing system 100
include
processing compensation and pensions, medical research, medical records
processing, etc.
[0071] The document processing system 100 can be configured to review
clinical
information provided with the request 102 including one or more of the message
104
and the documents 106 to determine compensation and pension benefits. The
guidelines 194 including requirements to be met for compensation and benefits
are
37
Date Recue/Date Received 2021-05-11
retrieved. The responsive data 196 is extracted from the information provided
by the
message and one or more of the external data sources 150. Based on the
threshold(s) met by the responsive data 196, the compensation and benefits for
a
candidate can be provided in the recommendation 140.
[0072] The document
processing system 100 can be employed for Medical
Record Processing at the Centers for Medicare and Medicaid Services, Military
Health System, etc., and for Risk Adjustment Data Validation in a use case.
The
document processing system 100 can be configured to review medical records for
processing for example, at Centers for Medicare and Medicaid Services (CMS),
Military Health System, etc. Furthermore, the document processing system 100
can
also be employed for Risk Adjustment Data Validation (RADV). The request 102
for
review of the medical records is received by the document processing system
100
and the medical records can be either received with the request 102 or may be
accessed from the external data sources 150 based on the information in the
request
102. The guidelines 194 for reviewing the medical records are retrieved and
the
responsive data 196 is extracted using the plurality of ML models 138 as
described
herein. The medical records may be reviewed to determine if they meet the
requirements in the guidelines 194 based on the approval thresholds as
discussed
herein.
[0073] Non-
clinical uses for the document processing system 100 can include,
functions such as procurement, customer engagement, etc.
38
Date Recue/Date Received 2021-05-11
[0074] The document processing system 100 can be configured to review
documents and contractual terms to recommend buying decisions. The request 102
can include information such as product lists, prices, etc. The guidelines 194
can
include contractual terms and the data extractor 128 extracts the responsive
data
196 from the request 102. If the responsive data 196 matches the requirements
set
forth by the contractual terms (i.e., the guidelines 194) as indicated by the
approval
threshold 452, then a recommendation to buy the products can be generated by
the
output generator 142. If the responsive data 196 fails to match the
requirements set
forth by the guidelines 194, then a recommendation against buying the products
can
be generated by the output generator 142.
[0075] The document processing system 100 can be configured with voice-
to-
text APIs so that the request 102 may not only be received in textual/document
format but may also be received as a voice message. NLP processing can be
implemented on the text extracted from the voice messages for processing of
customer benefit requests and questions.
[0076] Figure 11 illustrates a computer system 1100 that may be used to
implement the document processing system 100. More particularly, computing
machines such as desktops, laptops, smartphones, tablets, wearables which may
be used to generate or access the non-editable files corresponding to
unstructured
documents and their component documents may have the structure of the computer
system 1100. The computer system 1100 may include additional components not
shown and that some of the components described may be removed and/or
39
Date Recue/Date Received 2021-05-11
modified. In another example, the computer system 1100 can be implemented on
external-cloud platforms such as, but not limited to, Amazon Web Services,
AZURE
cloud or internal corporate cloud computing clusters, or organizational
computing
resources, etc.
[0077] The computer system 1100 includes processor(s) 1102, such as a
central
processing unit, ASIC or other type of processing circuit, input/output
devices 1112,
such as a display, mouse keyboard, etc., a network interface 1104, such as a
Local
Area Network (LAN), a wireless 802.11x LAN, a 3G, 4G or 5G, a mobile WAN or a
VViMax WAN, and a computer readable storage medium 1106. Each of these
components may be operatively coupled to a bus 1108. The computer readable
storage medium 1106 may be any suitable medium which participates in providing
instructions to the processor(s) 1102 for execution. For example, the computer
readable storage medium 1106 may be non-transitory or non-volatile medium,
such
as a magnetic disk or solid-state non-volatile memory or volatile medium such
as
RAM. The instructions or modules stored on the computer readable medium 1106
may include machine readable instructions 1164 executed by the processor(s)
1102
to perform the methods and functions of the document processing system 100.
[0078] The document processing system 100 may be implemented as software
stored on a non-transitory computer readable medium with processor-executable
instructions executed by one or more processors. For example, the computer
readable medium 1106 may store an operating system 1162, such as MAC OS, MS
WINDOWS, UNIX, or LINUX, and code or machine readable instructions 1164 for
Date Recue/Date Received 2021-05-11
the document processing system 100. The operating system 1162 may be a multi-
user, multiprocessing, multitasking, multithreading, real-time and the like.
For
example, during runtime, the operating system 1162 is running and the code for
the
document processing system 100 is executed by the processor(s) 1102.
[0079] The computer system 1100 may include a data storage 1110, which may
include non-volatile data storage. The data storage 1110 stores any data used
by
the document processing system 100. The data storage 1110 may be used to store
real-time data associated with the processes executed by the document
processing
system 100 such as the received requests, the various automatic document
processing tasks to be executed, the data 108 initially extracted from the
requests,
the ML models 138, the responsive data 196, the recommendations and the
letters
that are automatically generated and the like.
[0080] The network interface 1104 connects the computer system 1100 to
internal systems for example, via a LAN. Also, the network interface 1104 may
connect the computer system 1100 to the Internet. For example, the computer
system 1100 may connect to web browsers and other external applications and
systems via the network interface 1104.
[0081] What has been described and illustrated herein is an example
along with
some of its variations. The terms, descriptions and figures used herein are
set forth
by way of illustration only and are not meant as limitations. Many variations
are
possible within the spirit and scope of the subject matter, which is intended
to be
defined by the following claims and their equivalents.
41
Date Recue/Date Received 2021-05-11
