Process Mining With Generative AI for Unstructured Action Extraction

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional technologies are limited to extracting actions from structured data and cannot effectively process unstructured data, such as emails and chats, to identify and extract activities and tasks, relying on machine learning for intent identification only.

Innovation Solution

Implementing an extraction engine with generative artificial intelligence models to automatically process unstructured data, converting actions into activities or tasks within a process mining and discovery automation framework.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional machine learning techniques are used to process unstructured data, then intent identification can be achieved, but substantial action extraction cannot be performed

Engineering Contradiction:

Improveintent identification accuracyVSAvoidaction extraction capability

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent replaces conventional machine learning techniques with generative AI models (large language models) to process unstructured data. This substitution enables the system to not only identify intent but also extract substantial actions, activities, and tasks from unstructured communications, thereby resolving the limitation where ML could only perform intent identification without actionable extraction.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Productivity

If conventional process mining techniques are used, then actions can be extracted from structured data, but unstructured data cannot be processed

Engineering Contradiction:

Improveaction extraction from structured dataVSAvoidunstructured data processing capability

Core Design Contradiction:

ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent implements a universal extraction engine that can handle both structured and unstructured data formats. The system uses generative AI models to process diverse data types including emails, chats, comments, and other unstructured communications, while maintaining the ability to process structured data. This multi-functionality resolves the contradiction by enabling the same system to extract actions from both data types effectively.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Measurement precision

If only structured data is processed, then extraction accuracy is maintained, but the scope of extractable activities is limited

Engineering Contradiction:

Improveextraction accuracyVSAvoidrange of process activities

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent employs generative AI models with natural language processing capabilities to analyze unstructured data while maintaining extraction accuracy. The large language models understand context, semantics, and nuances in unstructured communications, enabling accurate identification of actions, activities, and tasks that would be impossible with conventional structured data processing alone, thus expanding the range of extractable process activities.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP4632598A1Process mining and discovery automation for extracting activities and tasks out of unstructured data

Publication Date: 2025.10.15 UIPATH INC

Figure 1
Figure 2
Figure 3

AI summary

A method is provided. The method is executed by an extraction engine implemented as a computer program within a computing environment. The extraction engine executes action and task mining on unstructured data. The method includes receiving a communication including unstructured data defining an action and automatically processing the communication by utilizing at least one generative artificial intelligence (AI) model to extract details of the action being performed in the unstructured data. The method includes automatically converting the action into an activity or a task of a process associated with the communication.