# US 2018/0322462 A1 · Model building architecture and smart routing of work items

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2018/0322462 A1
- Document ID: US20180322462A1
- Status: Published application
- Kind: A1
- Application number: 15/674,379
- Filed: 2017-08-10
- Published: 2018-11-08
- Earliest claimed: 2017-05-04
- Assignees: ServiceNow, Inc.
- Applicants: ServiceNow, Inc.
- Inventors: Baskar Jayaraman; Debashish Chatterjee; Kannan Govindarajan; Aniruddha Thakur
- Web page: https://patentgrep.com/patent/US20180322462A1

## CPC

- G06Q 10/103: {Workflow collaboration or project management}
- G06F 40/174: Form filling; Merging
- G06N 5/04: Inference or reasoning models
- G06N 20/00: Machine learning
- H04L 67/10: in which an application is distributed across nodes in the network (software deployment G06F8/60; multiprogramming arrangements G06F9/46)
- G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M)

## Abstract

Systems and methods for using a mathematical model based on historical information to automatically schedule and monitor work flows are disclosed. Prediction methods that use some variables to predict unknown or future values of other variables may assist in reducing manual intervention when addressing incident reports or other task-based work items. For example, work items that are expected to conform to a supervised model built from historical customer information. Given a collection of records in a training set, each record contains a set of attributes with one of the attributes being the class. If a model can be found for the class attribute as a function of the values of the other attributes, then previously unseen records may be assigned a class as accurately as possible based on the model. A test data set is used to determine model accuracy prior to allowing general use of the model.

## Claims

### Claim 1

A cloud-based computer system, comprising:
a memory partition; and
one or more network interfaces communicatively coupled to one or more processing units and to the one or more memory partitions;
wherein the one or more memory partitions comprise computer instructions that when executed by the one or more processing units cause the cloud-based computer system to provide at least one customer instance and a shared machine learning service,
wherein the at least one customer instance is communicatively coupled, via the one or more network interfaces, to a remotely executing client application and configured to:
obtain information defining parameters regarding historical data pertaining to one or more completed incident reports; and
provide at least a portion of the obtained information to the shared machine learning service; and
wherein the shared machine learning service is configured to:
obtain historical data from the at least one customer instance representing attributes of a set of resolved historical incident reports, the historical data comprising both structured and unstructured data values;
process the structured data values to create a first mathematical representation of attributes defined by the structured data;
parse the unstructured data to create generated structured data;
process the generated structured data to create a second mathematical representation of terms; and
analyze the first mathematical representation and the second mathematical representation to create a model sufficient to predict attributes of a newly created incident report.

### Claim 2 (depends on 1)

The cloud-based computer system of claim 1, wherein the machine learning service is further configured to include a machine learning scheduler configured to receive requests from at least two customer instances and invoke at least two machine learning trainer instances configured to execute independently and concurrently.

### Claim 3 (depends on 2)

The cloud-based computer system of claim 2, wherein each machine learning trainer instance purges all historical data after creation of a first model and prior to obtaining historical data for a second model.

### Claim 4 (depends on 1)

The cloud-based computer system of claim 1, wherein the machine learning service is further configured to retrain the model utilizing information regarding changes to previously predicted attributes.

### Claim 5 (depends on 1)

The cloud-based computer system of claim 1, wherein the machine learning service is further configured to:
partition the obtained historical data to create at least a preparation dataset and a separate test dataset prior to creating the model;
create the model using the preparation dataset; and
test the model using the test dataset.

### Claim 6 (depends on 1)

The cloud-based computer system of claim 1, wherein the shared machine learning service is further configured to publish the model to the at least one customer instance.

### Claim 7 (depends on 6)

The cloud-based computer system of claim 6, wherein the shared machine learning service is further configured to schedule a machine learning trainer instance based on a geographical location of the at least one customer instance.

### Claim 8 (depends on 1)

The cloud-based computer system of claim 1, wherein the machine learning service is further configured to be idempotent.

### Claim 9 (depends on 1)

The cloud-based computer system of claim 1, wherein the machine learning service is further configured to comprise restartable training and prediction functions.

### Claim 10 (depends on 1)

The cloud-based computer system of claim 1, wherein the machine learning service and the customer instance are configured for full instance redundancy.

### Claim 11 (depends on 1)

The cloud-based computer system of claim 1, wherein the machine learning service is configured to provide a training feature utilizing a first portion of cloud infrastructure and a prediction feature utilizing a second portion of cloud infrastructure, wherein the first portion and second portion maintain independence from each other.

### Claim 12

A method of creating a prediction model based on historical incident report data, the method comprising:
receiving a request, at a shared machine language service executing in a cloud-based architecture, to schedule creation of a prediction mode;
invoking a machine learning training function;
providing historical data obtained from a first customer instance to the machine learning training function, the historical data representing completed incident reports;
processing the historical data to create pre-processed model input data;
transforming the pre-processed model input data using transformation techniques to create transformed model data;
partitioning the transformed model data into training, tuning, and testing data sets; and
creating a target matrix from the training data set, the target matrix comprising N_gram values associated with a target identifier as indicated by the extracted historical data.

### Claim 13 (depends on 12)

The method of claim 12, wherein the historical data comprises data obtained from a customer instance representative of a defined time frame.

### Claim 14 (depends on 11)

The method of claim 11, wherein processing the historical data comprises removing junk characters or redundant information.

### Claim 15 (depends on 11)

The method of claim 11, wherein processing the historical data comprises adjusting different references to a common item to a consistent reference to the common item.

### Claim 16 (depends on 11)

The method of claim 11, wherein the transformation techniques include keyword extraction.

### Claim 17

A method of routing a work item through a work flow, the method comprising:
obtaining information pertaining to one or more attributes of an in-progress work item; and
correlating the obtained information with a corresponding one or more attributes of a plurality of completed work items using a predictive model to automatically identify a proposed progression of in-progress work item through the work flow,
wherein progression through the work flow comprises a plurality of distinct paths as determined by the one or more attributes.

### Claim 18 (depends on 17)

The method of claim 17, wherein the in-progress work item comprises an incident report.

### Claim 19 (depends on 18)

The method of claim 18, wherein the incident report has been defined by an end-user and the proposed progression comprises automatically associating a priority, a category, or an assignment with the incident report.

### Claim 20 (depends on 18)

The method of claim 18, wherein the incident report has had at least one corrective action applied without success and the progression through the work flow comprises escalating a priority of the incident report or re-assignment of the incident report based on a similar historical incident report represented in the model as successfully resolved.
