Beyond Named Entities: LICR - An Empirical Study of LLM-based Implicit Concept Recognition

Fuente: Zenodo
Saved in:
Bibliographic Details
Main Authors: Liu, Qizhuang, Claude, Opus 4.5
Format: Recurso digital
Language:English
Published: Zenodo 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866901118647271424
author Liu, Qizhuang
Claude, Opus 4.5
author_facet Liu, Qizhuang
Claude, Opus 4.5
contents <p>This study validates Large Language Models' capability to recognize implicit concepts—methodologies, principles, and patterns that texts embody but do not explicitly name. Through five cross-domain test cases (mathematical proof, algorithm description, economic phenomenon, game theory scenario, quantum physics experiment), we demonstrate that LLMs achieve 100% recall on core implicit concept recognition with 96% overall accuracy. However, LLMs cannot reliably output Wikidata Q-identifiers. We propose LICR (LLM-based Implicit Concept Recognition), a hybrid architecture combining LLM semantic understanding with Wikidata API precise mapping, providing a solution for implicit concept recognition that extends beyond traditional entity linking.</p> <p>Key contributions:<br>1. First formal definition of Implicit Concept Recognition Problem (ICRP)<br>2. Empirical validation of LLM implicit concept recognition capability<br>3. Identification of LLM limitations in arbitrary identifier memory<br>4. Proposal of LLM-API hybrid architecture (LICR)</p> <p>This work establishes prior art for the LICR methodology and architecture.</p>
format Recurso digital
id zenodo_https___doi_org_10_5281_zenodo_18211198
institution Zenodo
language eng
publishDate 2026
publisher Zenodo
record_format zenodo
spellingShingle Beyond Named Entities: LICR - An Empirical Study of LLM-based Implicit Concept Recognition
Liu, Qizhuang
Claude, Opus 4.5
Implicit Concept Recognition
Entity Linking
Large Language Models
Wikidata
Knowledge Graph
LICR
Natural Language Processing
Prior Art
<p>This study validates Large Language Models' capability to recognize implicit concepts—methodologies, principles, and patterns that texts embody but do not explicitly name. Through five cross-domain test cases (mathematical proof, algorithm description, economic phenomenon, game theory scenario, quantum physics experiment), we demonstrate that LLMs achieve 100% recall on core implicit concept recognition with 96% overall accuracy. However, LLMs cannot reliably output Wikidata Q-identifiers. We propose LICR (LLM-based Implicit Concept Recognition), a hybrid architecture combining LLM semantic understanding with Wikidata API precise mapping, providing a solution for implicit concept recognition that extends beyond traditional entity linking.</p> <p>Key contributions:<br>1. First formal definition of Implicit Concept Recognition Problem (ICRP)<br>2. Empirical validation of LLM implicit concept recognition capability<br>3. Identification of LLM limitations in arbitrary identifier memory<br>4. Proposal of LLM-API hybrid architecture (LICR)</p> <p>This work establishes prior art for the LICR methodology and architecture.</p>
title Beyond Named Entities: LICR - An Empirical Study of LLM-based Implicit Concept Recognition
topic Implicit Concept Recognition
Entity Linking
Large Language Models
Wikidata
Knowledge Graph
LICR
Natural Language Processing
Prior Art
url https://doi.org/10.5281/zenodo.18211198