Advances in Language Models and Knowledge Graphs

The field of natural language processing is moving towards integrating large language models (LLMs) with knowledge graphs to enhance their factual knowledge and improve their performance on various tasks. Recent research has focused on developing methods to materialize LLM knowledge into knowledge bases, synthesizing conversational data to improve LLMs' conversational capabilities, and leveraging knowledge graphs to enhance LLM-based recommendation systems. Another area of interest is the development of techniques for information extraction, relation extraction, and link prediction, which can be used to discover new knowledge and relationships from large corpora of text. Notably, the use of Bayesian optimization and multi-label contrastive learning has shown promising results in improving the performance and efficiency of these systems. Some noteworthy papers include: GPTKB v1.5, which introduces a massive knowledge base for exploring factual LLM knowledge. DocTalk, which presents a novel approach to synthesizing conversational data from existing text corpora. KERAG_R, which proposes a knowledge-enhanced retrieval-augmented generation model for recommendation. Topic Modeling and Link-Prediction for Material Property Discovery, which presents an AI-driven hierarchical link prediction framework for discovering new materials. SCoRE, which introduces a streamlined corpus-based relation extraction system using multi-label contrastive learning and Bayesian kNN.

Sources

GPTKB v1.5: A Massive Knowledge Base for Exploring Factual LLM Knowledge

DocTalk: Scalable Graph-based Dialogue Synthesis for Enhancing LLM Conversational Capabilities

KERAG_R: Knowledge-Enhanced Retrieval-Augmented Generation for Recommendation

DocIE@XLLM25: In-Context Learning for Information Extraction using Fully Synthetic Demonstrations

Topic Modeling and Link-Prediction for Material Property Discovery

FuDoBa: Fusing Document and Knowledge Graph-based Representations with Bayesian Optimisation

SCoRE: Streamlined Corpus-based Relation Extraction using Multi-Label Contrastive Learning and Bayesian kNN

Extracting ORR Catalyst Information for Fuel Cell from Scientific Literature

Built with on top of