A multimodal knowledge-driven framework that supports question answering on standard knowledge named BEST-KAG (Knowledge-Augmented Generation for Building Engineering STandards), which consistently outperforms multiple mainstream LLMs in terms of Expert evaluation, and metrics including BLEU, and ROUGE.
Abstract
Construction standards are critical for building safety and sustainability. Existing standard application workflows rely on keyword-based document retrieval and manual cross-clause interpretation, which cannot reliably support multi-clause reasoning, multimodal knowledge utilization, or traceable clause-level evidence linkage. To address these limitations, this study develops a multimodal knowledge-driven framework that supports question answering on standard knowledge named BEST-KAG (Knowledge-Augmented Generation for Building Engineering STandards). The framework introduces 1) a multimodal knowledge graph (MKG) for unified representation of document hierarchy and heterogeneous standard knowledge with various connections, 2) a rule-LLM hybrid knowledge construction pipeline for scalable multimodal knowledge extraction, creating a large MAG with 251 building engineering standards, 171,652 nodes and 310,914 edges, and 3) a graph-retrieval-based knowledge-augmented generation architecture for clause-grounded and traceable question answering. Experiments demonstrate that BEST-KAG consistently outperforms multiple mainstream LLMs in terms of Expert evaluation, and metrics including BLEU, and ROUGE, with the best improvement up to 74.01% compared to the baselines.
Results indicate that integrating multi-source domain knowledge with relation-preserved retrieval and attribute-supported filtering provides more focused and inspectable evidence, thereby supporting more accurate complex material question answering.
Peize Li, Xi Guo, Nan Yin et al.· Electronics· 0 citations
GraphQAG effectively supports users in identifying knowledge coverage gaps, examining generated QA pairs, and refining the QA pair set through graph-based interactions, demonstrating the usefulness of combining knowledge graphs, LLM-based generation, and visual analytics for producing more comprehensive and trustworthy QA pairs from long documents.
Yize Li, Ruiqi Yu, Tianya Pan et al.· arXiv.org· 0 citations
This study proposes EQAS (Empowered Question-Answering System), a hybrid framework designed to support context-aware question-answering and intelligent feedback generation in domain-specific knowledge environments. EQAS integrates fine-tuned transformer-based models, instruction-guided large language models, domain-specific knowledge graphs, and LangChain-based vector retrieval to improve contextual relevance, response quality, and feedback consistency. To evaluate the proposed framework, a benchmark dataset consisting of 10,000 real-world question–answer pairs was constructed from authentic user interactions and domain-related information resources. Experimental evaluation across established transformer-based architectures and recent instruction-tuned large language models showed that the Llama-3.3-70B-Instruct baseline achieved the highest standalone QA performance (F1: 77.42; EM: 44.60), while the EQAS transformer-based configuration achieved an F1 score of 75.48 and an Exact Match score of 41.80. A controlled ablation analysis using Llama-3.3-70B-Instruct as the fixed QA backbone further showed that incorporating knowledge graph enhancement increased the F1 score from 77.42 to 79.93 and the Exact Match score from 44.60 to 46.82, corresponding to absolute improvements of 2.51 and 2.22 points, respectively. A complementary human-centered evaluation of 1000 generative responses by three NLP researchers yielded an overall quality score of 4.34/5 across correctness, clarity, sufficiency, and helpfulness, with an overall Krippendorff’s α of 0.80. Furthermore, the framework provides context-sensitive explanatory feedback that can support user understanding and knowledge acquisition during information-seeking interactions. The findings suggest that EQAS offers a scalable solution for intelligent question-answering, feedback support, and knowledge assistance in complex information environments. The proposed framework highlights the potential of combining large language models with structured knowledge representations to support context-aware question-answering and explanatory feedback generation in domain-specific information environments.
Efficient semantic information processing and multi-hop knowledge reasoning have become essential technologies for intelligent information services and next-generation networked systems. To address inaccurate semantic understanding caused by short or ambiguous queries and insufficient reasoning capability under fragmented knowledge structures, this study proposes an intelligent question answering framework that tightly integrates Transformer-based semantic encoding with graph attention reasoning. The proposed architecture employs DeBERTa-v3-base and conditional random fields for entity recognition, combines dual-tower vector retrieval with cross-encoder reranking for semantic disambiguation, and constructs k-hop knowledge subgraphs enhanced by multi-layer Graph Attention Networks to achieve relation-aware information propagation and evidence aggregation. A multi-task joint optimization strategy incorporating adaptive gradient normalization, adjacency reconstruction regularization, and negative sampling is further introduced to improve long-tail generalization and reasoning robustness. Neo4j graph storage and FAISS vector indexing enable near real-time retrieval and scalable deployment. Experimental evaluation demonstrates high semantic understanding accuracy, interpretable multi-hop reasoning capability, and stable performance under short and colloquial queries, with superior Exact Match and path reasoning accuracy compared with baseline models. Beyond educational applications, the proposed framework provides an effective methodology for semantic information fusion, distributed knowledge reasoning, intelligent query processing, and adaptive decision support, offering valuable references for communication-enabled information systems, networked knowledge services, and intelligent information infrastructures related to Electromagnetic Waves, Antennas and Propagation engineering applications.
KGCaRe is proposed, a hybrid approach that combines neural retrieval with symbolic reasoning over LLM-generated KGs that consistently outperforms existing baselines, including Vanilla LLM, Code Prompt, Text Prompt, Think-on-Graph, Vanilla RAG, and HybridContextQA.
Ghanshyam Verma, Sima Sarkar, Devishree Pillai et al.· 0 citations
Experiments show that precision improves by filtering invalid candidates, while recall is preserved due to retaining candidates whose constraints are not explicitly violated, and a three-valued constraint semantics that avoids incorrect rejections under open-world assumptions.
E. Kitzelmann· Deutsche Jahrestagung für Kü...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
