arxiv_ai 92% Match Research Paper Researchers in QA and knowledge graphs,Developers of enterprise search solutions,Data scientists working with structured knowledge 2 weeks ago

Interpretable Question Answering with Knowledge Graphs

graph-neural-networks › knowledge-graphs

📄 Abstract

Abstract: This paper presents a question answering system that operates exclusively on a knowledge graph retrieval without relying on retrieval augmented generation (RAG) with large language models (LLMs). Instead, a small paraphraser model is used to paraphrase the entity relationship edges retrieved from querying the knowledge graph. The proposed pipeline is divided into two main stages. The first stage involves pre-processing a document to generate sets of question-answer (QA) pairs. The second stage converts these QAs into a knowledge graph from which graph-based retrieval is performed using embeddings and fuzzy techniques. The graph is queried, re-ranked, and paraphrased to generate a final answer. This work includes an evaluation using LLM-as-a-judge on the CRAG benchmark, which resulted in accuracies of 71.9% and 54.4% using LLAMA-3.2 and GPT-3.5-Turbo, respectively.

Authors (4)

Kartikeya Aneja

Manasvi Srivastava

Subhayan Das

Nagender Aneja

Submitted

October 22, 2025

arXiv Category

cs.CL

International Semantic Intelligence Conference (ISIC), Germany, 2025

arXiv PDF

Key Contributions

Presents an interpretable question answering system that operates solely on knowledge graphs, avoiding retrieval-augmented generation (RAG) with large language models (LLMs). It uses a paraphraser model for entity-relationship edges and employs graph-based retrieval with embeddings and fuzzy techniques. This approach offers a more transparent and potentially efficient alternative for knowledge-based QA.

Business Value

Enables the development of more transparent and explainable QA systems, crucial for regulated industries or applications where understanding the source of an answer is paramount. It also offers a potentially more cost-effective alternative to large LLM-based solutions.

Paper Metadata

Innovation Type

System Design and Algorithmic

Deployment Feasibility

Moderate. Requires building and maintaining a knowledge graph. The system components are modular and can be implemented with existing tools.

Limitations Addressed

Black-box nature of RAG-based LLM QA systems,Computational cost and potential for hallucination in large LLMs,Need for interpretable answers in knowledge-intensive domains

Technical Tags

Question AnsweringKnowledge GraphsGraph RetrievalInterpretabilityLLM AlternativesParaphrasingEmbeddingsFuzzy Matching

Research Topics

Information RetrievalKnowledge RepresentationNatural Language ProcessingGraph Databases

Methods & Architectures

Knowledge Graph QueryingGraph-based RetrievalEmbeddingsFuzzy TechniquesParaphrasing ModelLLM-as-a-judge evaluation Small Paraphraser ModelEmbedding Models

Applications & Tasks

Knowledge Management Enterprise Search Question Answering Systems Interpretable QAReducing reliance on large LLMs for QAQA over structured knowledge Answering questions using knowledge graphsProviding explainable answersBuilding QA systems without full RAG

Datasets & Benchmarks

Datasets

CRAG benchmark

Benchmarks

CRAG benchmark (71.9% with LLAMA-3.2, 54.4% with GPT-3.5-Turbo using LLM-as-a-judge)

Accuracy

Related Fields

Knowledge GraphsQuestion AnsweringInformation RetrievalGraph DatabasesExplainable AI (XAI)

Keywords

Question AnsweringKnowledge GraphGraph RetrievalInterpretabilityLLM AlternativeRAGEmbeddingsFuzzy MatchingParaphrasingStructured DataInformation Extraction

Academic Context

#Information Retrieval#Knowledge Representation#Natural Language Processing#Graph Databases

Technology Stack

Frameworks & Libraries

Graph database libraries (e.g., Neo4j drivers)Embedding libraries

Programming Languages

Python

Data Processing Tools

Knowledge graph construction tools

Commercial Potential

Potential Products

Interpretable QA enginesKnowledge graph-powered search toolsDomain-specific expert systems

Target Industries

HealthcareFinanceLegalCustomer SupportResearch

Use Case Examples

Answering complex medical queries with traceable evidenceProviding financial advice with clear justificationsBuilding internal knowledge bases for large organizations

Competitive Edge

Offers a distinct advantage in interpretability and reduced reliance on large, potentially opaque LLMs for knowledge-based QA tasks.

Market Opportunity

Growing demand for explainable AI and efficient knowledge management solutions.

Revenue Models

Licensing of the QA systemconsulting services for KG development and integration.

Resource Requirements

Compute Needs

Moderate, depends on the size of the knowledge graph and the complexity of queries.

Data Requirements

A well-structured knowledge graph.

Deployment Constraints

Requires a curated and maintained knowledge graph. Performance depends heavily on the quality and coverage of the KG.

Scalability

Scalability depends on the underlying graph database technology and the efficiency of the retrieval and paraphrasing modules.

Production Readiness

Maturity Level

Research

Time to Market

1-3 years for integration into specific applications.

Patent Potential

Moderate, for specific techniques in graph retrieval and paraphrasing for QA.

View Full Paper Back to Papers