Idea
Query suggestion platform improving local-life service search relevance and efficiency with geography-aware LLM and optimized inference.
Research Paper
Core Innovation
This paper introduces LocalSUG, which integrates geographic grounding into LLM-based query suggestion via city-aware candidate mining. It addresses exposure bias with a beam-search-driven GRPO training algorithm and reduces online latency through quality-aware beam acceleration and vocabulary pruning, outperforming traditional multi-stage systems.
Why It Matters
Local-life service platforms face challenges in providing relevant query suggestions due to geographic diversity and long-tail demand. LocalSUG reduces user effort and search latency while increasing engagement by delivering more accurate, location-aware suggestions. This improves user satisfaction and business metrics at scale, transforming local search workflows.
Market Size (TAM)
$10–20B TAM for search and recommendation platforms; $2–5B SAM from local-life and e-commerce service providers. Driven by increasing demand for personalized, location-aware search and efficiency in online query processing.
Potential Customers & Pain Points
- Local-life service platforms – Need relevant location-specific query suggestions
- E-commerce platforms – Need to reduce no-result searches
- Search engine providers – Need to optimize inference latency and suggestion quality.
Business Model
SaaS platform licensing or API subscription for local-life service providers and e-commerce platforms, with tiered pricing based on query volume and feature set.
Competitive Landscape
- Google Autocomplete
- Bing Autosuggest
- Alibaba Query Suggestion
- Baidu Search Suggestion
Implementation Challenges
- Integration complexity with existing local-life platforms
- Balancing inference speed with suggestion quality
- Data privacy concerns with geographic data usage
Validation Strategy
- Conduct large-scale A/B testing on partner local-life platforms
- Measure CTR improvement and reduction in no-result rates
- Optimize latency and resource usage in production environments
Research Paper Overview
LocalSUG: Geography-Aware LLM for Query Suggestion in Local-Life Services
Summary
LocalSUG is an LLM-based query suggestion framework designed for local-life service platforms to improve user search experience by generating geographically grounded candidate queries, reducing exposure bias, and minimizing online latency. It enhances click-through rates and lowers no-result occurrences through city-aware candidate mining, a beam-search-driven training algorithm, and efficient inference optimizations.