Knowledge Graphs and Ontologies: Adding Knowledge to Keyword Search (Algolia)
This entry appears in a curated reading list section on RAG (Retrieval-Augmented Generation). The listed source link points to:
Key points
- Topic: Adding knowledge via knowledge graphs and ontologies to keyword/keyword-based search, positioned at the intersection of information retrieval and large-scale search/recommendation systems.
- Motivation (as stated in the template): Traditional pipelines treat retrieval, ranking, and generation separately; LLM-era applications require natural language interaction, multi-hop reasoning, and fresh knowledge, motivating tighter coupling of external knowledge sources with generative models.
- Canonical pipeline described: problem formulation → model/system design → training or construction process → inference pipeline, with modules such as retrievers, rerankers, planners, memory, and tool interfaces.
- Inference strategies covered in the template: single-step retrieval, iterative retrieval, parallel sub-queries, early stopping, and compute budget control.
- Evaluation framing: datasets like MS MARCO, BEIR, Natural Questions; metrics like nDCG@10, MRR, Recall@k; baselines like BM25, dense retrieval, cross-encoder reranking, and retrieval-free LLMs. The post explicitly warns that specific numeric results must be verified against the original paper — it reports none itself.
- A Survey of Graph Retrieval-Augmented Generation for Customized Large Language Models (arXiv 2501.13958)
- A Survey on Retrieval-Augmented Text Generation for Large Language Models (arXiv 2404.10981)
- Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG (arXiv 2501.09136)
- RAFT: Adapting Language Model to Domain Specific RAG (Jul 2024)
- RAG vs. GraphRAG: A Systematic Evaluation and Key Insights (arXiv 2502.11371)
- AutoKnow: Self-driving knowledge collection for products of thousands of stores
- IR — Information Retrieval
- RAG — Retrieval-Augmented Generation
- LTR — Learning to Rank
- nDCG — Normalized Discounted Cumulative Gain
- Agentic Search — modeling search as sequential decision-making and tool invocation
- Gen-IR — Generative Information Retrieval
Engineering checklist (from the post)
| Area | Question | Suggestion | |------|----------|-----------| | Data | PII in training/index? Version management? | Partitioned indexes, anonymization, rollback-able embedding versions | | Latency | p99 budget? How many retrieval steps? | Cascaded retrieval + early stop, cache hot queries, async reranking | | Quality | Does offline gain translate to online CTR/satisfaction? | Interleaving experiments, human audits, citation checks | | Safety | Does open retrieval introduce poisoning/bias? | Source allowlists, adversarial detection, output filtering | | Cost | Token and GPU cost per query? | Small-model routing, distillation, hybrid sparse+dense retrieval |
Related cross-references listed in the post
Glossary (from the post)
Caveats
The post is largely a bibliographic template: several summary fields (one-line summary, technical route, engineering trade-offs) are unfilled placeholders, and the abstract line is a repeat of the title. The most concrete factual anchor is the linked Pinterest Interest Taxonomy engineering article on building a knowledge graph management system for content understanding.