SIGIR 2026

Status at time of writing (2026-07-24): the conference is on its final day — Friday is the workshop day. Written from the published program, not attendance.

Main-conference days run six parallel oral sessions (5–6 papers each) in two blocks, 11:00–12:30 and 13:30–15:00, around posters and coffee. 51 oral sessions total across recommendation, retrieval, ranking, RAG, multimodal learning, and fairness, plus industry sessions.

Talk titles were not on the program page at time of writing — fill in from the program PDF or the proceedings front matter.

1. Bridging Personalization and AI: From RAG to Agent
2. LLM Personalization: Foundations, Breakthroughs, and Frontiers
3. Multi-Agentic Recommender Systems: Foundations, Perspectives, and Lessons from Large Scale Deployments in eCommerce
4. Retrieve, Rerank, Answer, Experiment: Hands-On IR Research with PyTerrier
5. Temporal Information Retrieval and Extraction: From Foundations to RAG
6. MANILA26: Information Retrieval for Climate Change Impact

#4 (PyTerrier) is the one with direct carry-over here — a reproducible retrieve/rerank/evaluate harness is exactly the missing piece in the pocket-es work, which has a ranking function but no standing relevance evaluation.

1. AgentSearch: Indexing, Retrieval, and Ranking of AI Agents
2. Second Workshop on Explainability in Information Retrieval
3. SCAI'26 — 10th Workshop on Search-Oriented Conversational AI
4. ReNeuIR — Fifth Workshop on Reaching Efficiency in Neural Information Retrieval
5. ECOM26 — SIGIR 2026 Workshop on eCommerce
6. Second Workshop on Evaluation of Multimodal Generation
7. LLM-UP — LLM-powered User Profiling for Search and Recommendation
8. VulGen'26 — Vulnerabilities in Generative Systems for Information Retrieval
9. JEDI — Justice, Emancipation, Democracy, and Information Access: Resisting Corporate and Authoritarian Capture of Information Access Platforms

Three of these bear directly on work here: ReNeuIR (efficiency — the whole premise of a client-side index), AgentSearch (agents as retrievable objects, which is the retrieval side of the agent-discovery documents under .well-known/), and VulGen (attacks on generative retrieval, adjacent to the bot/compliance thread).

Retrieval, in-house

pocket-es — client-side BM25 over this site · spec · SIPs, n-grams and phrase search · search UX spec · contracts.

RAG / agents

llm-agent-frameworks · amazon-bedrock-rag-workshop · agentic-2026.

Reasoning models feeding retrieval

ACM TechTalk — reasoning models to agents.

Same body, same season

SIGGRAPH 2026 (Jul 19–23, overlapped by four days) · ACM AI Leadership Summit 2026.

Neural retrieval has displaced lexical baselines for practical corpora

a ReNeuIR result where BM25 (or BM25 + cheap reranking) matches a neural stack within noise at a fraction of the cost on a small corpus — i.e. the pocket-es regime.

Agents are usefully modelled as retrievable, rankable objects (AgentSearch)

an agent-selection task where ranking by description is no better than a flat registry lookup or a hand-written router.

RAG improves answer quality over long-context prompting

a task where feeding the whole corpus into a long-context model beats retrieval at equal cost.

(no term)

• [ ] Pull keynote titles + abstracts from the program PDF
• [ ] ReNeuIR proceedings: any efficiency result that transfers to a browser-side index (pruning, quantised postings, SIP-style skipping)
• [ ] PyTerrier tutorial materials — candidate harness for a standing relevance eval over the site corpus (~811 docs)
• [ ] AgentSearch papers vs the .well-known/ agent-discovery documents here
• [ ] Best-paper and test-of-time awards once posted
• [ ] Check whether SIGIR 2027 host/dates are announced