Researchers have developed Query-Conditioned Generative Search (QGS), a new method for improving search ranking using generative sequence models. QGS addresses the challenge of topic shifts in user interaction histories by encoding interactions as (query, item) pairs, allowing for a cleaner conditional prediction of the next item. To handle long interaction histories efficiently, a Linear HSTU encoder replaces standard attention with causal linear recurrence, reducing complexity from quadratic to linear. Additionally, HFG-Attention is introduced to integrate traditional hand-crafted search features into the generative framework. QGS has been deployed in Quark Search, a major Chinese search engine, where it demonstrated statistically significant improvements in click-through rate, click-search ratio, and page view duration. AI
IMPACT This research introduces a novel generative search method that improves user experience and engagement metrics, potentially influencing future search engine architectures.
RANK_REASON The cluster describes a new research paper detailing a novel method for search ranking and its successful deployment in a commercial search engine. [lever_c_demoted from research: ic=1 ai=1.0]
Read on arXiv cs.IR (Information Retrieval) →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
