CTR-Guided Generative Query Suggestion in Conversational Search (EMNLP 2025 Industry Track)

Forum topic · 小凯 · 2026-07-05

Summary

This forum entry indexes an EMNLP 2025 Industry Track paper titled "CTR-Guided Generative Query Suggestion in Conversational Search," published in the ACL Anthology. The work addresses query suggestion in conversational search systems, using click-through rate (CTR) signals to guide generative models when proposing follow-up queries. It sits at the intersection of conversational information retrieval, generative query reformulation, and industrial ranking systems, where user engagement signals such as clicks inform what suggestions are worth generating. The surrounding forum post situates the paper within the broader landscape of conversational search research, including surveys, agentic retrieval systems, and generative query reformulation methods like ConvGQR. It also discusses typical system architectures (retriever, reranker, planner, generator), evaluation protocols (nDCG, MRR, Recall@k, latency), and engineering constraints such as cost, safety, and online deployment. Readers should consult the original PDF for exact quantitative results, as this entry is based on publicly available metadata.

CTR-Guided Generative Query Suggestion in Conversational Search (EMNLP 2025)

Overview

This entry catalogues an industry-track paper accepted at EMNLP 2025, published in the ACL Anthology:

Title: CTR-Guided Generative Query Suggestion in Conversational Search
Venue: EMNLP 2025 Industry Track (ACL)
Source: https://aclanthology.org/2025.emnlp-industry.178/
Topic area: Conversational Search

The paper explores generative query suggestion for conversational search, where a system proposes follow-up queries during multi-turn interactions. The distinguishing idea, per the title, is using click-through rate (CTR) signals from user behavior logs to guide the generation process, so that suggested queries are not only contextually relevant but also aligned with what users are likely to click.

> Note: The detailed abstract content in the source forum post was a placeholder. For exact methods, experiments, and quantitative results, refer to the official ACL Anthology page above.

Context Within Conversational Search

Query suggestion sits in a lineage of conversational IR research that includes:

Generative query reformulation (e.g., ConvGQR)
LLM-based retrieval adaptation (e.g., ChatRetriever)
Collaborative and agentic search systems (e.g., CoSearchAgent, agentic conversational search with reinforcement learning)
Broad surveys of conversational search systems and applications

The typical conversational search pipeline encodes query/document/context representations, then chains retrieval, reranking, and generation modules, with feedback signals (clicks, sessions) increasingly used as supervision.

Why CTR Signals Matter

Using CTR to guide query suggestion reflects a common industrial insight: offline relevance metrics (nDCG, MRR) do not always predict user satisfaction. Click logs provide an implicit preference signal that can steer generative models toward suggestions that are both contextually appropriate and behaviorally validated in production traffic.

Practical Considerations for Deployment

Engineering teams evaluating this line of work should consider:

Latency: p99 budgets and the number of retrieval/generation steps
Data quality: PII handling, log de-biasing, and position-bias in CTR signals
Safety: open-ended generation risks, source filtering
Evaluation: interleaving experiments and human audits to verify offline gains translate to online engagement

Related Entries

A Survey of Conversational Search (ACM, 2025)
Engineering Conversational Search Systems: A Review of Applications
Agentic Conversational Search with Contextualized Reasoning
ChatRetriever: Adapting LLMs for Generalized and Robust Retrieval
CoSearchAgent: A Lightweight Collaborative Search Agent
ConvGQR: Generative Query Reformulation for Conversational Search