Nanopublications LDF server

Matches in Nanopublications for { ?s <http://www.w3.org/2000/01/rdf-schema#comment> ?o ?g. }

• 20964471.2024.2429847 comment "Important practical advantage of DGGS over raster workflows. Land-use mapping often involves classifying based on text attributes (e.g., 'Recreation Reserve'), which is difficult with numeric-only raster formats. DGGS integration with databases like PostgreSQL enables full-text search capabilities that significantly improve classification flexibility." assertion.
• 20964471.2024.2429847 comment "The paper presents a case study of land-use classification for the Northland region of New Zealand. This region was chosen as the demonstration area for the DGGS-based land-use mapping approach, integrating 48 different geospatial layers using the H3 DGGS at resolution 13. Northland is located in the northern part of New Zealand's North Island, bounded approximately by 172.0-174.5°E longitude and 34.4-36.2°S latitude." assertion.
• W5B4-WK93 comment "Multi-temporal, thematic classification of New Zealand's land cover with 33 mainland classes. Produced at 1:50,000 scale with time steps from 1996/97 to 2018/19. Used as one of 48 input geospatial layers for the Northland land-use mapping demonstration. The paper notes this dataset would be appropriately indexed at H3 resolution 10 or 11 (2,100 m² to 15,000 m² average hexagon area). Note: v5.0 is deprecated; v6.0 is now current." assertion.
• v1.1.1 comment "Synthetic geospatial datasets for benchmarking DGGS performance against vector and raster workflows. Includes 500 random vector coverages using Voronoi polygons and 10,000 NLM (neutral landscape model) raster landscapes. Data is deterministically reproducible from seeded random number generators." assertion.
• the-semantic-web comment "The paper that introduced the Semantic Web. A big inspiration to many, even though many things unfolded differently than envisaged in this paper." assertion.
• rio.10.e121887 comment "This quote captures the fundamental distributed nature of environmental research, where field work and data collection must happen across diverse geographic locations. It establishes the philosophical foundation for why research infrastructures like LifeWatch ERIC require distributed architecture." assertion.
• rio.10.e121887 comment "This explains the core motivation for creating LifeWatch ERIC - providing long-term sustainability and visibility for biodiversity research outputs that might otherwise be lost or remain isolated. It addresses a critical gap in research infrastructure coordination." assertion.
• rio.10.e121887 comment "This quote describes the dual funding model of LifeWatch ERIC, which balances national autonomy (in-kind contributions) with centralized coordination (cash contributions). This hybrid model is essential for distributed European research infrastructures." assertion.
• rio.10.e121887 comment "This provides concrete detail about the assessment process - 30 questions covering relevance, quality, and cost justification. This structured approach ensures systematic and transparent evaluation of in-kind contributions across all member countries." assertion.
• rio.10.e121887 comment "This acknowledges a practical limitation in peer review of diverse technical contributions. The authors suggest relying on traceability, visibility, scientific authority of contributors, and external expert consultation as quality indicators." assertion.
• rio.10.e121887 comment "This reflects the reality of evaluating diverse research contributions - trust-based governance is sometimes necessary when quantitative metrics are insufficient. This has implications for how distributed research infrastructures can practically operate." assertion.
• rio.10.e121887 comment "This establishes the expectation that in-kind contributions are not one-time donations but ongoing commitments requiring maintenance and user support. This sustainability requirement is crucial for long-term infrastructure viability." assertion.
• rio.10.e121887 comment "Belgium is one of the eight member countries of LifeWatch ERIC, contributing through in-kind contributions to the distributed research infrastructure." assertion.
• rio.10.e121887 comment "Bulgaria is one of the eight member countries of LifeWatch ERIC, contributing through in-kind contributions to the distributed research infrastructure." assertion.
• rio.10.e121887 comment "Italy is one of the eight member countries of LifeWatch ERIC. Italy hosts the LifeWatch ERIC Service Centre, one of the Common Facilities." assertion.
• rio.10.e121887 comment "Portugal is one of the eight member countries of LifeWatch ERIC, contributing through in-kind contributions to the distributed research infrastructure." assertion.
• rio.10.e121887 comment "Slovenia is one of the eight member countries of LifeWatch ERIC, contributing through in-kind contributions to the distributed research infrastructure." assertion.
• rio.10.e121887 comment "Spain is one of the eight member countries of LifeWatch ERIC. Spain hosts the Statutory Seat and ICT e-Infrastructure Technical Offices, which are Common Facilities." assertion.
• BDJ.12.e113943 comment "This establishes the long-term nature of the study (13 years from 2010-2023), making it valuable for understanding temporal patterns in herbivory-plant interactions and species conservation." assertion.
• BDJ.12.e113943 comment "The experimental design separates the effects of domestic livestock (sheep) from wild herbivores (rabbits), allowing assessment of their individual and combined impacts on plant communities." assertion.
• BDJ.12.e113943 comment "The protected status of A. europaeum highlights the conservation importance of understanding how herbivory management affects this endangered species." assertion.
• BDJ.12.e113943 comment "This morphological trait indicates evolutionary adaptation to grazing pressure, supporting the hypothesis that moderate herbivory may benefit rather than harm this species." assertion.
• BDJ.12.e113943 comment "Land use change, not herbivory, appears to be the primary threat to A. europaeum, contextualizing the importance of understanding herbivory effects for conservation management." assertion.
• BDJ.12.e113943 comment "Study site is within Amoladeras Nature Reserve, a hunting refuge and zoological reserve within Cabo de Gata-Níjar Natural Park. The area has limestone soils, belongs to the Murcian-Almerian chorological province, thermo-Mediterranean semi-arid-arid belt, with 200mm annual precipitation and 19°C mean annual temperature." assertion.
• BDJ.12.e113943 comment "Almería Province is the primary distribution area for A. europaeum in Spain, containing the study site and multiple populations of this endangered species." assertion.
• BDJ.12.e113943 comment "Western Morocco represents the African portion of the Ibero-Maghreb distribution range of A. europaeum, complementing the Spanish populations." assertion.
• jpjhuu comment "Darwin Core Archive containing 13 years (2010-2023) of annual monitoring data on herbivory effects on the endangered species Androcymbium europaeum and its associated plant communities. The dataset includes: 1583 sampling events (18 plots × 13 years with nested quadrats and transects), 4011 occurrence records of 100 plant taxa from 31 families, and 6922 extended measurement records including species cover, vegetation cover, species richness, and Shannon diversity index. Data collected using randomized block design with three treatments: herbivory by sheep and rabbits (G+R+), exclusion of domestic livestock (G-R+), and exclusion of both rabbits and domestic livestock (G-R-)." assertion.
• assertion comment "Was created by claude-ai-agent and not me!" assertion.
• AIAgentDrivenKGConstructionFramework comment "This framework leverages LLM-powered agents across three stages: ontology creation and expansion, ontology refinement, and knowledge graph population. Its primary goal is to automate the construction of product-specific KGs from unstructured text, directly performing tasks like entity discovery and relation extraction to build and populate the graph." assertion.
• KgLlmAblationFramework comment "This framework serves as a baseline variant of the KG-LLM approach. It converts multi-hop KG paths into natural language prompts but explicitly removes instructions, textualized IDs, and Chain-of-Thought reasoning from the expected response. It uses instruction fine-tuning to train LLMs for multi-hop link and relation prediction as a comparative method." assertion.
• KgLlmAblationFramework comment "This method is a variant of the KG-LLM Framework, specifically designed with a simplified "ablation knowledge prompt." Unlike the full KG-LLM prompt, it removes explicit instructions, textualized IDs, and CoT reasoning from the expected response. It is introduced and evaluated by the authors as a baseline within their proposed approach to demonstrate the effectiveness of the advanced prompting strategies in the full KG-LLM Framework for multi-hop link prediction." assertion.
• KgLlmFramework comment "This framework converts multi-hop KG paths into Chain-of-Thought natural language prompts with instructions. These prompts are then used to instruction fine-tune LLMs, enabling them to perform multi-hop link prediction and relation prediction tasks within knowledge graphs, significantly leveraging LLMs to enhance KG completion." assertion.
• KgLlmFramework comment "The KG-LLM Framework is a novel method that converts structured KG data (multi-hop paths) into natural language Chain-of-Thought (CoT) prompts. These prompts are then used to instruction fine-tune LLMs to enhance multi-hop link prediction and relation prediction performance in KGs. The method leverages LLMs' generative and reasoning capabilities to solve KG completion tasks." assertion.
• KgLlmFramework comment "The KG-LLM framework is a novel method that converts structured knowledge graph paths into natural language Chain-of-Thought (CoT) prompts. These prompts are then used to instruction fine-tune Large Language Models (LLMs) and integrate In-Context Learning (ICL) for enhancing multi-hop link and relation prediction in KGs. This method leverages LLMs to improve knowledge graph completion tasks." assertion.
• OdaObservationDrivenAgent comment "ODA is a novel AI agent framework designed for KG-centric tasks, employing an iterative "observation, action, and reflection" cycle. It deeply integrates LLMs and KGs by allowing the LLM to act as an agent that interacts with the KG to conduct reasoning, where KG observations guide the LLM's decision-making and memory updates." assertion.
• KGAgent comment "KG-Agent is an autonomous LLM-based agent framework designed to improve complex reasoning over KGs. It integrates a fine-tuned LLM as a planner, a multifunctional toolbox for KG interaction, a KG-based executor, and knowledge memory, using an iterative mechanism for tool selection and memory updates to conduct reasoning. This fits 'SynergizedReasoning' as it treats LLMs as agents interacting with KGs for complex reasoning." assertion.
• GenerateOnGraph comment "GoG is a training-free method designed for Incomplete Knowledge Graph Question Answering (IKGQA). It treats the LLM as both an agent to explore KGs and as a knowledge source to generate new factual triples when existing KG knowledge is insufficient. The method's Thinking-Searching-Generating framework enables the LLM to bridge knowledge gaps and answer questions that conventional KGQA methods fail on due to KG incompleteness." assertion.
• GenerateOnGraph comment "GoG is a training-free method for Incomplete Knowledge Graph Question Answering (IKGQA). It employs a Thinking-Searching-Generating framework, treating the LLM as both an agent to explore the KG and as a knowledge source to generate missing factual triples. This synergistic approach combines LLM's reasoning and internal knowledge with external KG facts to answer complex questions where the KG is incomplete." assertion.
• GenerateOnGraph comment "GoG proposes a "Thinking-Searching-Generating" framework where the LLM acts as an agent to explore KGs and also as a KG to generate new factual triples using its internal knowledge. This synergy of LLM and KG capabilities enables effective reasoning for Question Answering over incomplete Knowledge Graphs by dynamically augmenting missing information." assertion.
• hybridInstructionPretrainingMechanism comment "This method is introduced during the Continuous Pre-training (CPT) phase of PediatricsGPT. It leverages data from PedCorpus (which incorporates knowledge graph resources) by assembling instruction data into completion forms and assimilating them into plain texts. This mechanism explicitly aims to bridge knowledge discrepancies and enhance the LLM's medical domain adaptation during pre-training." assertion.
• knowledgeEnhancedPromptForCptCorpus comment "This method uses specific prompts to transform structured instruction data from the KG-inclusive PedCorpus dataset into comprehensive medical knowledge texts. These texts are integrated into the PedCorpus-CPT dataset. This process directly enriches the continuous pre-training corpus for the LLM, thereby improving its knowledge expression during pre-training." assertion.
• rolePlayingDrivenInstructionBuildingRule comment "This rule describes a method within PedCorpus construction that uses GPT-4 API to generate specialized pediatric instructions. It explicitly extracts knowledge from pediatric textbooks, guidelines, and knowledge graphs to create high-quality, professional, and humanistic instruction data. This KG-enhanced data then feeds into the LLM's pre-training and fine-tuning phases, thereby enhancing the LLM's knowledge." assertion.
• KGRank comment "KG-Rank is an augmented LLM framework that integrates a medical Knowledge Graph (KG) with a pipeline of entity extraction, relation retrieval, and novel ranking/re-ranking techniques (Similarity, Answer Expansion, MMR, MedCPT re-ranking). Its purpose is to enhance the factual consistency and reliability of LLMs during the inference stage for long-form medical question answering." assertion.
• KGRank comment "KG-Rank is an augmented LLM framework that integrates a medical knowledge graph (KG) with ranking and re-ranking techniques to improve the factuality of long-form question answering (QA). It identifies medical entities, retrieves related KG triples, and applies various ranking methods (Similarity, Answer Expansion, MMR) followed by re-ranking (MedCPT) to refine the information provided to the LLM during inference for answer generation. The primary goal is to enhance LLM performance by leveraging external knowledge from KGs at inference time." assertion.
• CogMG comment "CogMG is a collaborative augmentation framework where an LLM, acting as an agent, decomposes queries, identifies knowledge gaps in KGs, completes missing facts using its internal parameters, and facilitates the evolution of the KG through verified additions. This mutual enhancement directly improves LLM's factual accuracy in QA by leveraging the updated KG, while also enriching the KG with LLM-generated and verified knowledge, thereby enhancing reasoning across both components." assertion.
• CogMG comment "CogMG is a collaborative augmentation framework where an LLM acts as an agent to interact with a Knowledge Graph. It involves the LLM decomposing queries for the KG, completing missing knowledge triples using its internal parameters, and then verifying and integrating these new facts into the KG. This continuous feedback loop represents a synergistic reasoning process, where the LLM addresses knowledge gaps in the KG and leverages the enriched KG to improve its factualness in QA." assertion.
• ChatEA comment "ChatEA is a novel framework that leverages LLMs' background knowledge and multi-step reasoning abilities to significantly improve Entity Alignment, a crucial task in Knowledge Graph construction and integration. It achieves this by translating KG structures into an LLM-understandable "code" format and employing a two-stage iterative reasoning process for alignment decisions." assertion.
• LLMAlign comment "LLM-Align is a three-stage framework for Entity Alignment (EA) that leverages LLMs. It uses heuristic methods for attribute and relation selection to create informative prompts and a multi-round voting mechanism to mitigate LLM hallucinations and positional bias, aiming to improve the accuracy of matching entities across KGs. Entity alignment is a crucial step in KG integration, falling under KG construction." assertion.
• LLMAlign comment "LLM-Align is a three-stage framework that leverages Large Language Models (LLMs) to infer entity alignments across Knowledge Graphs (KGs). It employs heuristic methods for selecting informative attributes and relations to construct specific prompts for LLMs and utilizes a multi-round voting mechanism to mitigate hallucination and positional bias, thereby enhancing the accuracy and reliability of KG completion by identifying equivalence facts." assertion.
• LLMEA comment "LLMEA is proposed as the first framework to integrate knowledge from KGs and LLMs for entity alignment. It enhances KG construction by using LLMs to generate virtual equivalent entities via knowledgeable prompts and to perform final alignment predictions through multi-choice question answering, thereby leveraging LLM's semantic knowledge and inference capabilities to improve a core KG construction task." assertion.
• GraphCheck comment "GraphCheck is an LLM-based verifier that incorporates lightweight graph signals alongside instruction-style prompting. The graph signals are used during the LLM's inference to enhance its verification capabilities by providing structured insights." assertion.
• KgCraft comment "KG-CRAFT is a novel method that leverages knowledge graphs to formulate structured contrastive questions. These KG-derived questions guide LLMs during the inference stage of fact-checking, improving their ability to assess claim veracity by providing distilled, evidence-based summaries. The primary goal is to enhance the LLM's performance in automated fact-checking through structured reasoning based on KG information." assertion.
• AiAgentDrivenFrameworkForAutomatedProductKgConstruction comment "This framework leverages LLM-powered agents to fully automate the end-to-end process of constructing product knowledge graphs directly from unstructured text. It operates in three main stages: ontology creation and expansion, ontology refinement, and knowledge graph population, all aimed at improving and automating KG construction tasks." assertion.
• ODAObservationDrivenAgent comment "ODA is a novel AI agent framework designed for KG-centric tasks that deeply integrates LLM and KG reasoning. It operates through a cyclical Observation-Action-Reflection paradigm, where the LLM acts as an agent interacting with the KG. The core contribution is an observation module that autonomously processes KG knowledge (using a recursive D-turn observe strategy) to generate reasoning patterns, which then guides the LLM's action and reflection steps, enabling synergistic reasoning for complex KG-related questions." assertion.
• KgAgent comment "KG-Agent is an autonomous LLM-based agent framework designed to synergize LLMs and KGs for complex reasoning tasks over KGs. It treats the LLM as a planner that actively interacts with the KG through a specialized multifunctional toolbox and an iterative memory-updating mechanism. This aligns perfectly with the 'treating LLMs as agents to interact with the KGs to conduct reasoning' aspect of SynergizedReasoning." assertion.
• KgAgent comment "KG-Agent is an autonomous LLM-based agent framework that integrates an instruction-tuned LLM, a multifunctional toolbox, a KG-based executor, and knowledge memory. It enables LLMs to actively make decisions and iteratively interact with KGs for complex reasoning. This aligns with the 'SynergizedReasoning' subcategory, which includes treating LLMs as agents to interact with KGs to conduct reasoning." assertion.
• KgRank comment "KG-Rank is an augmented LLM framework that integrates a medical Knowledge Graph with novel ranking and re-ranking techniques. It identifies entities, retrieves relevant KG triples, and applies methods like Similarity Ranking, Answer Expansion Ranking, and MMR Ranking, followed by MedCPT-based re-ranking to provide relevant and precise information to LLMs during inference. The primary goal is to improve the factual consistency and reliability of LLM-generated long-form answers in medical Question Answering." assertion.
• CogMg comment "CogMG is a unified framework where the LLM acts as an agent, using its planning and reasoning capabilities to interact with the KG. It actively decomposes queries, processes results, and orchestrates the "Graph Evolution" by identifying and completing missing knowledge triples, which are then used to update the KG. This continuous, agent-driven interaction and mutual updating represent a synergistic reasoning process where the LLM leverages the KG and its own knowledge to answer questions and improve the KG's knowledge base." assertion.
• COGATELECTRA comment "CO-GAT (ELECTRA) applies graph attention over scientific evidence, with ELECTRA (a pre-trained language model) as its encoder. This method enhances the LLM's understanding and reasoning by integrating graph structures into its evidence processing during inference." assertion.
• FactLLaMA comment "FactL-LaMA is an LLM-based method that enhances fact-checking by augmenting LLMs with external knowledge sources during inference. This external knowledge (often KG-based in this context) helps improve the LLM's accuracy in verification tasks." assertion.
• IKA comment "IKA is a method that uses example graphs (Knowledge Graphs) to enhance LLM capabilities for claim verification and explanation. The KGs provide structured context during the LLM's inference to improve its fact-checking performance." assertion.
• KGCraft comment "KG-CRAFT is a novel method that enhances LLM capabilities for automated fact-checking. It leverages KGs by first using LLMs to construct a KG from claims and reports. Then, the KG structure is used to formulate contrastive questions, which guide the LLM's reasoning process during the inference stage to synthesize evidence and assess claim veracity." assertion.
• AiAgentDrivenFrameworkForAutomatedProductKnowledgeGraphConstruction comment "This method introduces a novel, fully automated framework that leverages LLM-powered agents to construct product knowledge graphs. It performs three key stages: ontology creation and expansion, ontology refinement, and knowledge graph population. The primary goal is to enhance and automate the KG construction task by using LLMs." assertion.
• kgLlmAblationFramework comment "This is an ablation variant of the KG-LLM framework, where the knowledge prompt structure is simplified by removing explicit instructions, textualized IDs, and Chain-of-Thought reasoning from the expected response. It is explicitly designed and evaluated as a baseline within the same LLM-augmented KG completion approach." assertion.
• kgLlmFramework comment "The KG-LLM framework converts multi-hop knowledge graph paths into natural language Chain-of-Thought prompts. These prompts are then used for instruction fine-tuning of Large Language Models (LLMs) to perform multi-hop link and relation prediction, which are tasks aimed at completing missing information within Knowledge Graphs." assertion.
• AiAgentDrivenProductKgConstructionFramework comment "This framework leverages LLMs through a multi-agent system to automate the entire process of constructing product knowledge graphs. It involves three stages: ontology creation and expansion, ontology refinement, and knowledge graph population, directly addressing the entity discovery, coreference resolution, and relation extraction tasks inherent in KG construction from unstructured text." assertion.
• ObservationDrivenAgentODA comment "ODA is a novel AI agent framework designed for KG-centric tasks. It integrates LLMs and KGs through a cyclical paradigm of observation, action, and reflection, enabling the LLM to leverage autonomous reasoning patterns from the KG. This framework creates a synergistic reasoning process where the LLM acts as an agent interacting with and being guided by the KG's knowledge to enhance problem-solving." assertion.
• PediatricsGPTTrainingPipeline comment "The PediatricsGPT method proposes a systematic training pipeline for an LLM assistant. A key component of this pipeline, the "hybrid instruction pre-training mechanism" within the Continuous Pre-Training phase, explicitly integrates knowledge graph resources into the PedCorpus-CPT dataset. This integration directly enhances the LLM's foundational medical knowledge and domain adaptation during pre-training, ultimately improving its performance in pediatric applications." assertion.
• LlmAlign comment "LLM-Align is a novel three-stage framework that leverages LLMs for Entity Alignment (EA) in Knowledge Graphs. It employs heuristic methods for selecting informative attributes and relations from KGs to create prompts for LLMs, and introduces a multi-round voting mechanism to enhance the reliability of LLM-inferred alignments. This method directly addresses the KG construction task by improving entity matching across KGs." assertion.
• KGCRAFT comment "KG-CRAFT is a novel framework that synergizes LLMs and KGs for automated fact-checking. It leverages LLMs for knowledge graph construction, then uses the KG structure to formulate contrastive questions that guide LLMs to generate evidence-based answers and summaries for veracity assessment. This process exemplifies LLMs acting as agents interacting with KGs to conduct structured reasoning." assertion.
• KGo1 comment "KG-o1 is a four-stage framework designed to enhance the intrinsic multi-hop reasoning abilities of LLMs. It leverages KGs to filter entities, generate logical paths, construct complex QA datasets for supervised fine-tuning (SFT) to simulate long-term thinking, and applies a Self-improved Adaptive DPO strategy to refine the LLMs' reasoning, ultimately improving LLM performance during inference for multi-hop question answering." assertion.
• QuestionAwareKnowledgeGraphPrompting comment "This method enhances LLM performance in Multiple Choice Question Answering (MCQA) by generating query-adaptive soft prompts from Knowledge Graphs (KGs) during the inference stage. It integrates question embeddings into GNN aggregation (QNA) to assess KG relevance and uses global attention across answer options (GTP) to enrich soft prompt completeness, without requiring LLM fine-tuning." assertion.
• LecKg comment "The LEC-KG framework introduces a novel bidirectional collaborative approach for knowledge graph construction, integrating LLMs and KGEs for mutual enhancement. It employs KGE to provide structure-aware feedback for refining LLM extractions via evidence-guided Chain-of-Thought reasoning, while LLM-validated triples progressively improve KGE representations. This iterative feedback loop combines the distinct reasoning capabilities of both LLMs and KGEs to enhance the overall KG construction process." assertion.
• ROG comment "ROG is a framework designed to improve complex logical reasoning over knowledge graphs using LLMs. It achieves this by decomposing FOL queries, retrieving query-relevant subgraphs, and employing LLM-based chain-of-thought reasoning to answer these queries step-by-step. The method directly enhances a KG task (answering complex logical queries) by leveraging LLM capabilities." assertion.
• ROG comment "ROG is a retrieval-augmented framework that combines query-aware neighborhood retrieval from Knowledge Graphs with Large Language Model chain-of-thought reasoning to answer complex First-Order Logic queries. It explicitly decomposes multi-operator queries, grounds each step in compact, retrieved KG evidence, and caches intermediate answer sets. This approach treats the LLM as an agent interacting with the KG to conduct complex, multi-step logical reasoning, thus synergizing both components for enhanced reasoning capabilities." assertion.
• KgTraces comment "KG-TRACES is a unified framework that trains LLMs with explicit supervision from Knowledge Graph-derived paths and attribution-aware reasoning processes. This enables the LLM to perform structured, explainable, and attributable reasoning by internalizing KG-aligned reasoning patterns and generating explanations that are grounded in KG facts or inferred knowledge. This approach synergizes LLMs and KGs by teaching the LLM to effectively conduct reasoning with both." assertion.
• KGRAGFramework comment "This method proposes a novel framework that integrates a Knowledge Graph (KG) with Retrieval-Augmented Generation (RAG) to enhance Large Language Model (LLM) performance in the telecommunications domain. The KG is used during the inference stage to dynamically provide relevant, structured, and up-to-date domain-specific knowledge to the LLM, improving its accuracy and contextual understanding for tasks like question answering and summarization without retraining." assertion.
• LLMAidedEntityExtraction comment "This method uses an LLM to extract key entities and their relationships from unstructured and semi-structured telecom-specific data sources. The LLM processes documents, tokenizes them into segments, and applies predefined prompts to identify named entities and their types (e.g., protocol, metric, component) along with their semantic context, which are then used to construct the domain-specific Knowledge Graph." assertion.
• Kicgpt comment "KICGPT is a framework that integrates an LLM with a triple-based KGC retriever to improve Knowledge Graph Completion performance, particularly for long-tail entities. The LLM re-ranks candidate entities provided by the retriever, leveraging its internal knowledge and external KG context to enhance the KG task." assertion.
• KnowledgePrompt comment "Knowledge Prompt is a novel in-context learning strategy proposed within KICGPT that encodes KG structural knowledge into demonstrations. It guides the LLM to effectively conduct reasoning for the KGC task by providing relevant KG context, thus facilitating a synergized reasoning process." assertion.
• TextSelfAlignment comment "Text Self-Alignment is a method that uses the LLM via in-context learning to transform raw and obscure KG relation descriptions into more natural and comprehensible text. This process synergizes knowledge representation by making KG information more accessible and aligned with the LLM's understanding." assertion.
• KgBasedRandomWalkReasoning comment "This method leverages a knowledge graph's structured causal relationships by performing a multi-step random walk to extract contextually relevant information. This extracted information, including verbalized triples and randomly selected subsequent nodes, is integrated into LLM prompts during the inference stage to enhance the LLM's reasoning abilities and performance in commonsense question answering tasks without requiring retraining." assertion.
• PrivCompKG comment "PrivComp-KG is a novel, unified framework that combines LLM-extracted knowledge with a structured knowledge graph and its reasoning capabilities to verify privacy policy compliance. The LLM populates the KG with extracted relations between privacy policies and regulatory articles, enabling the KG to effectively represent the complex compliance landscape for subsequent inference via SWRL rules." assertion.
• RAGLLMComplianceFactExtraction comment "This method proposes a specific pipeline using a RAG-enabled LLM (Llama-7B with ChromaDB and a vector store of GDPR article chunks) and a tailored prompting strategy to identify relevant GDPR articles and segments aligning with privacy policy sections. This extracted information directly contributes to populating the PrivComp-KG, thus augmenting KG construction by providing extracted compliance facts." assertion.
• KgFit comment "KG-FIT is a novel framework that enhances Knowledge Graph Embeddings (KGE) by integrating open-world entity knowledge from Large Language Models (LLMs). It uses LLMs to guide hierarchical clustering of entities and refine the hierarchy, then fine-tunes KG embeddings using this LLM-derived hierarchical structure, textual embeddings, and standard link prediction objectives. The goal is to improve KG embeddings and their expressiveness for tasks like link prediction." assertion.
• linkKg comment "LINK-KG is a modular LLM-based framework designed to improve knowledge graph construction from complex legal texts. It leverages a three-stage, LLM-guided coreference resolution pipeline with a type-specific prompt cache and structured prompts for entity and relationship extraction. This method enhances KG tasks by addressing entity discovery, coreference resolution, and relation extraction." assertion.
• PerozziEtAl2025KGGraphToken comment "This method introduces a novel technique for directly integrating structured Knowledge Graph (KG) representations into frozen Large Language Models (LLMs). It extends the GraphToken framework to KGs by leveraging Knowledge Graph Embedding (KGE) models to generate structured tokens from KG information. These tokens are concatenated with the natural language query and fed into the frozen LLM at inference time, enhancing the LLM's reasoning capabilities without requiring LLM fine-tuning or prompt engineering." assertion.
• Dkgllm2025AdaptiveSemanticFusionAlgorithm comment "ASFA is the core innovation within the DKG-LLM framework, responsible for integrating Grok 3 outputs with the DKG, ensuring semantic consistency, and managing computational efficiency. It includes phases for data ingestion, semantic extraction, dynamic graph updates, and a dedicated 'Reasoning and Recommendation' phase utilizing Bayesian inference and optimization for diagnosis and treatment. ASFA facilitates the synergistic reasoning process by dynamically fusing information and refining parameters through feedback." assertion.
• Dkgllm2025DkgllmFramework comment "The DKG-LLM Framework integrates a Dynamic Knowledge Graph (DKG) with the Grok 3 LLM to provide medical diagnosis and personalized treatment recommendations. It addresses the LLM's shortcomings in transparent reasoning by leveraging the DKG for structured knowledge and the Adaptive Semantic Fusion Algorithm (ASFA) for dynamic updates and probabilistic reasoning. The framework's core aim is to enhance decision-making and diagnostic accuracy by synergistically combining both components." assertion.
• RAGFLARKOMultiStagePipeline comment "This method proposes a multi-stage Retrieval-Augmented Generation (RAG) pipeline to enhance the performance of LLMs in financial asset recommendations. It uses LLMs to select relevant entities and construct compact subgraphs from a Personal Knowledge Graph (PKG) and a Market Knowledge Graph (MKG) in a chained, sequential manner. The curated KG context is then fed to a financial LLM (FLARKO) during inference to improve recommendation quality, reduce hallucinations, and optimize context window usage." assertion.
• SAT comment "SAT (Structure-aware Alignment-Tuning) is a novel framework designed to enhance LLMs for Knowledge Graph Completion (KGC) tasks. It achieves this by introducing hierarchical knowledge alignment to bridge the representation gap between graph structures and natural language, and structural instruction tuning with a lightweight adapter to unify various KGC tasks. The primary goal is to improve the LLM's performance on KG completion tasks." assertion.
• ReaLM comment "ReaLM is a novel framework designed to enhance Knowledge Graph Completion (KGC) tasks by bridging the gap between continuous KG embeddings and discrete LLM token spaces. It achieves this by quantizing pretrained KG embeddings into compact code sequences, integrating them as learnable tokens within the LLM's vocabulary, and using ontology-guided class constraints to refine predictions. This method directly aims to improve LLM performance on KG tasks like link prediction and triple classification." assertion.
• lightprof comment "LightPROF is a Retrieve-Embed-Reason framework that uses a Knowledge Adapter to encode both textual and structural information from Knowledge Graphs into LLM-friendly soft prompts. These prompts are injected during the LLM's inference stage to enhance its reasoning capabilities on KGQA tasks, particularly for small-scale LLMs, without retraining the LLM itself. This approach directly improves LLM performance during inference through KG integration." assertion.
• LIKR comment "LIKR is a unified framework where an LLM is treated as a reasoner that outputs intuitive exploration strategies for a Knowledge Graph (KG). It integrates LLM's text-based output (user preferences) with KG embeddings into a hybrid reward function for a reinforcement learning agent. This agent then performs KG path reasoning to make recommendations, synergizing both LLM intuition and KG structure for enhanced reasoning capabilities." assertion.
• deductiveVerificationBeamSearch comment "Deductive-Verification Beam Search (DVBS) is a core component of FiDeLiS that constructs and validates reasoning paths step-by-step. It uses LLM-generated planning to guide the beam search and employs an LLM-based deductive verification mechanism (local and global checks) to ensure logical consistency and factual correctness, enabling synergized reasoning with LLMs and KGs." assertion.
• fideliS comment "FiDeLiS is a unified, training-free framework designed to improve the factual accuracy and reasoning efficiency of LLM responses in KGQA. It grounds LLM answers in verifiable reasoning steps from KGs by integrating the Path-RAG and Deductive-Verification Beam Search modules, representing a synergized approach to reasoning with both LLMs and KGs." assertion.
• pathRAG comment "Path-RAG is a retrieval-augmented generation module within the FiDeLiS framework that pre-selects a smaller candidate set of entities and relations from the KG for each beam search step. It combines LLM-generated keywords, semantic similarity, and KG structural connectivity to efficiently narrow the search space and support the overall synergized reasoning process." assertion.
• ChainOfKnowledgeCoK comment "This method introduces a comprehensive framework for integrating knowledge reasoning into LLMs by learning from Knowledge Graphs. It includes KG-based dataset construction (KNOWREASON) and a vanilla fine-tuning strategy, enabling LLMs to conduct a form of reasoning deeply informed by KG logic, thus creating a synergized reasoning capability." assertion.
• PathsOverGraphPoG comment "PoG is a novel method that enhances LLM reasoning by integrating knowledge reasoning paths from KGs during the inference stage. It tackles multi-hop and multi-entity questions through dynamic multi-hop path exploration and efficient pruning techniques, allowing LLMs to access factual knowledge for more faithful and interpretable outputs without retraining. The primary goal is to improve LLM reasoning performance for knowledge-intensive tasks like KGQA." assertion.
• PathsOverGraphPoGE comment "PoG-E is an introduced variant of PoG, designed to evaluate the impact of graph structure on KG-based LLM reasoning. It operates by randomly selecting one relation from each edge in the clustered question subgraph during the path exploration, making it a specific approach within the KGEnhancedLLMInference category to study the effect of structural information on LLM performance." assertion.

• first
• previous
• next