Integration of the Semantic Wiki Technologies with Large Language Models as a Technological Basis for Experience Acquisition from Natural Language Documents

Igor SINITSYN D. Sc. (Tech.), Prof., Corr. Memb. of the NAS of Ukraine

Julia ROGUSHINA Ph.D. (Phys. & Math.), Senior Res., Assoc. Prof.

Kostiantyn YURCHENKO Postgrad. Stud., Junior Res. Fellow

Yurii BOVA Postgrad. Stud., Eng.

The Institute of Software Systems

of the National Academy of Sciences of Ukraine

October 1, 2025

ПРИКЛАДНІ СИСТЕМИ ТА ТЕХНОЛОГІЇ В ІНФОРМАЦІЙНОМУ СУСПІЛЬСТВІ

​

1

Introduction

2

The Data Deluge

The modern digital era generates vast volumes of heterogeneous electronic documents, rich in natural language information but often unstructured. This exponential data accumulation presents significant challenges for intelligent analysis.

Analytical Imperative

Efficient intelligent analysis of these large, unstructured datasets is crucial. The goal is to extract pertinent information, satisfy user needs, and transform raw data into actionable knowledge.

Toward Knowledge & Experience

Effective analysis demands tools that facilitate the transition from raw data to structured knowledge. These tools must leverage existing experience, integrate with established knowledge sources, and support collaborative expert input.

Research Objectives

Define the class of Experience Acquisition from Documents (EAD) tasks

3

​

​

1

Integrate semantic technologies with LLMs for linguistic analysis

​

​

2

Develop models and methods enabling control over correctness of acquired knowledge

​

​

3

Create technological platform for transforming natural language documents into structured knowledge

​

​

4

Ensure explainability and reliability of AI-generated results

​

​

5

Proposed Approach: LLM-Semantic Wiki Integration

Solution Strategy: Integrate Semantic Technologies (specifically Semantic MediaWiki – SMW) with Generative AI (Large Language Models – LLMs).

4

Key Advantage

This integration combines the power of LLMs with the reliability and explainability of semantic analysis results.

LLM

Provide linguistic analysis of document content and Natural Language text generation.

Semantic Wiki

Acts as a verifiable knowledge base. It makes the new domain knowledge acquired by LLMs visible, understandable, and editable.

​

The "Linza" Technological Platform

Platform: "Linza" – designed for intelligent processing of unstructured documents to generate structured CIOs according to user requirements.

5

Methodology: Explainability and Trust

Methodology: The EAD task is divided into independent stages, and intermediate results are fixed using formalized semantic markup.

LLM Problem: High performance, but open issues regarding the reliability of results and the possibility of explaining the reasoning paths.

Linza's Solution - Applying verified knowledge structures (from ontology/SMW) to control the LLM's output generation.

6

Explainability: The user receives explanations that explicitly track and indicate which formalized content elements and logical rules were utilized in the result construction.

Outcome: This explicit explanation becomes an important component of trust in the system, especially in critical subject domains.

Human-in-the-Loop: Experts can supervise and validate each processing stage and its results to ensure factual accuracy and increased trust.

Ontological model of “Linza” ecosystem

7

Practical Application Examples

8

Intelligent Assistant for a PhD Student:

Personalized Set

Selection of lectures, textbooks, and courses according to research topics and interests.

Consideration of Interests

Takes into account the scientific interests of the postgraduate student and their supervisor, as well as language proficiency.

Open Access Documents

Most documents are open access, simplifying processing.

Chatbot Assistant

Provides explanations, finds relevant documents, and suggests information.

Retrieval and Analysis of Precedents in the Accumulated Experience of the Armed Forces of Ukraine��Problems:�Big number of documents��Low-formalized combat lessons� obtained by the AFU��Confidential information needs processing by local LLMs��Critical consequences of the low-quality outcomes

9

Practical Application Examples �

Practical Application Examples �

Constructing a target security profile for an information system:

Task: Creating certification documents based on example pairs and a large number of up-to-date standards.

Challenge: Large volume of regulatory documents and the need to construct RDs with a complex, formalized structure.

10

Conclusions: Advantages and Challenges

11

The integration of generative AI and semantic technologies provides the power of LLMs with explainable results.

Ontological Modeling

The basis of the terminology system for analysis and recommendations.

Semantic Markup

Specialized LLM for creating an integrated knowledge base.

Semantic Wiki-technologies

The basis for searching and verifying markup.

Intelligent Chatbot

Natural language dialogue for searching and clarifying needs.

Machine Learning

Algorithms for classifying experience and finding patterns.

Thanks to semantic Wikis, LLM operation ceases to be a "black box," ensuring transparency and reliability of decisions.

Future Work & Discussion

• Scalability of the platform across larger and more diverse document corpora.
• More complex knowledge models and rules of data transformation.
• Further optimization of LLM prompting and fine-tuning by domain-specific semantic feedback.
• Integration of advanced reasoning methods to enhance the rule generation module.

​

Thank you for your attention.

Questions?

12