UltRAG: a Universal Simple Scalable Recipe for Knowledge Graph RAG

This feed aggregates recent IR/LLM research (Apr 2026) focused on retrieval-augmented generation (RAG) for knowledge graphs, multimodal/graph-vector fusion, embedding design, continual tokenization for generative recommenders, cross-domain knowledge distillation, and automated RDF extraction from cloud logs. Key contributions include ULTRAG (LLMs + neural query executors for large-scale KGQA), calibrated score fusion for heterogeneous graph/vector retrieval, evaluations of single-vector vs multi-vector embeddings, drift-aware tokenizer updates (DACT), and an LLM-based pipeline for Log→RDF KG生成