Transformer Entity Analysis for Unlabeled Unstructured Data

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing systems struggle to effectively analyze and categorize unstructured data from IT infrastructure logs, medical records, and live chat transcripts due to their non-uniform nature and lack of labeling, making it difficult to extract operational insights.

Innovation Solution

A multi-machine learning model system combining a supervised transformer classifier-based model for 'crisp' entities and an unsupervised transformer embedding-based model for 'hazy' entities, employing hierarchical clustering and density-driven segregation to identify and filter relevant information from unstructured data.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If traditional NLP with unsupervised learning is used to cluster and analyze unstructured data, then the system can process data without requiring labels, but the analysis precision and ability to identify critical information deteriorates

Engineering Contradiction:

Improveability to process unstructured data without labelsVSAvoidanalysis precision

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The system segments the analysis process into two distinct stages: a supervised stage that identifies and extracts critical entities with high precision, and an unsupervised stage that clusters remaining data for broader patterns. This segmentation allows each stage to optimize for its specific purpose, resolving the contradiction between labeled and unlabeled processing capabilities.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The supervised model performs preliminary action by first identifying and extracting critical entities from the unstructured data before the unsupervised clustering stage processes the remaining data. This preliminary extraction ensures that critical information is captured with high precision while the unsupervised stage handles the bulk of the data efficiently.

Inventive Principle:

Principle #10Preliminary action

2Device complexity

If a single machine learning model is used to analyze unstructured data, then the system complexity is reduced, but the ability to handle both known and unknown entities deteriorates

Engineering Contradiction:

Improvesystem complexityVSAvoidability to handle known and unknown entities

Core Design Contradiction:

Device complexityVSAdaptability or versatility

Solution Approach 1:

The system merges two different machine learning approaches - supervised learning for known entities and unsupervised learning for unknown entities - into a unified multi-model system. This combination allows the system to handle both crisp and hazy entities effectively, achieving high adaptability while managing complexity through integration.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The multi-model system achieves universality by designing a system where the supervised model handles known entities and the unsupervised model handles unknown entities, making the overall system capable of addressing diverse entity types without requiring separate specialized systems for each case.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Productivity

If unstructured data is analyzed without categorization and labeling, then the data processing speed increases, but the ability to identify critical incidents and extract operational insights deteriorates

Engineering Contradiction:

Improvedata processing speedVSAvoidoperational insights

Core Design Contradiction:

ProductivityVSLoss of information

Solution Approach 1:

The supervised model performs preliminary action by rapidly identifying and extracting critical entities from the unstructured data stream, enabling the system to process data at high speed while capturing essential operational insights. This preliminary extraction ensures that critical information is not lost during the processing pipeline.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system extracts critical entities and operational insights from the unstructured data through the supervised model, separating the most important information from the bulk data. This extraction allows the system to maintain high processing speed while preserving essential insights, as the extracted critical information can be immediately acted upon.

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS20260037567A1Multi-machine learning model system for unstructured data

Publication Date: 2026.02.05 HEWLETT PACKARD ENTERPRISE DEV LP

AI summary

Systems and methods are provided for analyzing unstructured data using at least two machine learning models in a multi-machine learning model system, including (1) a supervised machine learning model that may be implemented as a transformer classifier-based entity recognition model operating on known entities (“crisp” entities), and (2) an unsupervised machine learning model that may be implemented as a transformer embedding-based model operating on unknown entities (“hazy” entities). The combination of the two models may execute a hierarchical and cascaded analysis of the input data that combines a clustering technique with a density-driven segregation of entities. Output of the multi-model system may help identify potential important information and non-relevant information to quickly examine critical incidents as well as possible non-relevant information.