CTR-Guided Generative Query Suggestion in Conversational Search (EMNLP 2025 Industry Track)
Overview
This entry indexes an academic paper from the EMNLP 2025 Industry Track, published via the ACL Anthology:
- Title: CTR-Guided Generative Query Suggestion in Conversational Search
- Venue: EMNLP 2025 (Industry Track), ACL
- Link: https://aclanthology.org/2025.emnlp-industry.178/
- Topic area: Conversational Search
- The paper targets query suggestion in conversational search, where a system must proactively propose follow-up queries across multi-turn interactions.
- Its distinguishing idea is the use of click-through rate (CTR) signals to guide the generative suggestion process, aligning suggested queries with both semantic relevance and observed user engagement behavior.
- The work sits at the intersection of large-scale search/recommendation systems and LLM-era conversational interaction, where traditional pipelines that treat retrieval, ranking, and generation separately struggle to meet user expectations for natural language, multi-turn experiences.
- A Survey of Conversational Search, Sep 2025, ACM
- Engineering Conversational Search Systems: A Review of Applications (arXiv:2407.00997)
- ConvGQR: Generative Query Reformulation for Conversational Search (arXiv:2305.15645)
- ChatRetriever: Adapting LLMs for Generalized and Robust Retrieval (arXiv:2404.13556)
- CoSearchAgent: A Lightweight Collaborative Search Agent (arXiv:2402.06360)
Key points
Context within conversational search
The accompanying discussion frames the paper within broader trends:
1. Architecture: Cascaded retrieval + reranking + generation remains the mainstream stack, but agentic paradigms increasingly treat retrieval policy itself (whether and how many times to retrieve) as a learnable decision. 2. Data: High-quality instruction data and click/session logs are both critical; synthetic data must guard against knowledge leakage and distribution shift. 3. Evaluation: The gap between offline metrics (nDCG, MRR, Recall@k) and online user satisfaction is widening; LLM-as-judge approaches need cross-validation against human evaluation. 4. Productization: Latency, cost, interpretability, and safety are hard constraints in industrial deployment and cannot be sacrificed purely for benchmark gains.
Related entries
Caveats
The forum post is largely a structured meta-analysis based on the paper's metadata rather than its full content. Specific experimental setups, datasets, and quantitative results are not reproduced here — readers should consult the original PDF at the ACL Anthology link above before citing numbers.
Glossary
| Term | Meaning | |------|---------| | CTR | Click-Through Rate — the ratio of users who click a suggested item | | IR | Information Retrieval | | RAG | Retrieval-Augmented Generation | | nDCG | Normalized Discounted Cumulative Gain, a ranking quality metric | | Agentic Search | Modeling search as sequential decision-making with tool use |