Query Auto-Completion Ranking via Earth Mover's Distance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing query auto-completion systems in eCommerce do not effectively account for dynamic changes in product information supply and demand in consumer-to-consumer online marketplaces, leading to inefficient product retrieval.
Innovation Solution
The method employs Earth Mover's Distance (EMD) matching to evaluate search terms based on costs and frequencies, constructing a cost matrix and optimizing term vectors to rank queries for product retrieval efficiency, considering both query-side demand and product-side supply constraints, thereby suggesting the most efficient search terms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional query auto-completion methods (MPK or basic learning-based approaches) are used, then implementation simplicity is maintained, but product retrieval efficiency deteriorates due to ignoring dynamic product information supply changes
Solution Approach 1:
The patent implements dynamic query auto-completion by continuously updating the supply-demand difference based on real-time product information changes. The system dynamically adjusts the QAC ranking by incorporating the time-varying supply-demand difference, making the previously static MPK approach adaptive to changing marketplace conditions. This resolves the contradiction by making the system dynamic without requiring complete redesign.
Solution Approach 2:
The patent introduces a feedback mechanism where the supply-demand difference is calculated based on product information changes and fed back into the query auto-completion ranking. This feedback loop allows the system to adjust recommendations based on actual marketplace dynamics, improving retrieval efficiency while maintaining a relatively simple architecture by reusing existing components with an added feedback layer.
2Measurement precision
If query auto-completion considers only query-side features (time, location, behavior), then implementation complexity is reduced, but retrieval accuracy deteriorates due to ignoring product supply dynamics
Solution Approach 1:
The patent merges query-side demand features with product-side supply features by calculating the supply-demand difference. Instead of treating them separately, the system combines both aspects into a unified metric that reflects the actual marketplace state. This merging improves evaluation accuracy by considering both perspectives without requiring entirely separate processing pipelines.
Solution Approach 2:
The patent segments the feature engineering into distinct components: query-side features (time, location, behavior) and product-side features (supply changes, product information updates). By segmenting the problem, the system can process each type of feature independently and then combine them through the supply-demand difference calculation, making the overall system more manageable and interpretable.
3Reliability
If static product information assumptions are made, then processing speed is improved, but retrieval relevance deteriorates in C2C marketplaces where product information changes frequently
Solution Approach 1:
The patent applies partial action by not requiring complete re-processing of all product information changes. Instead, it tracks and processes only the relevant supply-side changes that affect query auto-completion rankings. This selective approach maintains retrieval relevance by responding to important changes while avoiding the overhead of processing every possible update, thus preserving processing speed.
Data Source
AI summary
For query auto completion (QAC) in an eCommerce platform with constantly changing product supply and demand, a Query Mover's Distance (QMD) framework for ranking queries is used, which formulates QAC as an optimal transport problem balancing product demand and supply.