Environment-Aware Search Planning for E-Commerce
A novel approach known as Environment-Aware Search Planning (EASP) tackles the challenge of blindness-latency in industrial e-commerce searches. While Large Language Models (LLMs) excel in reasoning, they either rewrite queries without considering retrieval capabilities and current inventory or utilize deep search agents that result in delays of several seconds. EASP employs a Probe-then-Plan strategy: a lightweight Retrieval Probe reveals the retrieval snapshot, allowing a Planner to identify execution gaps and create grounded search plans. This methodology consists of three phases: Offline Data Synthesis, during which a Teacher Agent generates a variety of execution-validated plans.
Key facts
- EASP reformulates search planning as dynamic reasoning grounded in environmental reality.
- Probe-then-Plan uses a lightweight Retrieval Probe to expose the retrieval snapshot.
- The Planner diagnoses execution gaps and generates grounded search plans.
- The methodology has three stages: Offline Data Synthesis, Teacher Agent, and execution validation.
- LLM-based paradigms face a blindness-latency dilemma in industrial search.
- Query rewriting is agnostic to retrieval capabilities and real-time inventory.
- Deep search agents rely on iterative tool calls and reflection, incurring seconds of latency.
- Industrial sub-second budgets are incompatible with high-latency methods.
Entities
—