Set Expansion System Using Quality Weighting and Negative Seeds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing set expansion techniques, such as SEISA, perform poorly with noisy web data, particularly when expanding large concept sets, as they treat all web lists equally and do not account for negative seeds, leading to reduced precision and recall.
Innovation Solution
The proposed set expansion system improves precision and recall by weighting lists based on the presence of positive and negative seeds, applying a cutoff to eliminate low-quality lists, and using distributed processing techniques like Map Reduce to refine the candidate result set, allowing for multiple passes to achieve high-quality expansions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If SEISA treats each web-list as equal for set expansion, then the system is simple to implement, but the precision drops when expanding large concept sets due to noise from less popular entities
Solution Approach 1:
The patent applies local quality by assigning different weights to different web-lists based on their quality metrics (number of positive seeds, negative seeds, and entity count). Instead of treating all lists equally, high-quality lists with more positive seeds and fewer negative seeds receive higher weights, while low-quality lists are downweighted or filtered out. This localized quality adjustment resolves the contradiction by maintaining simplicity through automated weighting rather than complex manual curation, while simultaneously improving precision by reducing noise from low-quality lists.
2Productivity
If SEISA does not use negative seeds to filter entities, then the processing is faster and simpler, but bad entities are introduced into the expanded set reducing precision
Solution Approach 1:
The patent applies preliminary action by incorporating negative seed information during the candidate generation phase rather than requiring a separate filtering pass. The weighting scheme uses negative seeds to downweight lists containing bad entities before they are added to the expanded set. This preliminary filtering action maintains high precision while avoiding the need for slow post-processing rejection, thus resolving the contradiction between processing speed and precision.
3Measurement precision
If seed-specific and page-specific wrappers are used to reduce candidate set, then precision is improved, but the architectural complexity and system responsiveness deteriorate
Solution Approach 1:
The patent applies universality by creating a general-purpose web-list quality weighting system that works across different domains and seed types without requiring domain-specific wrappers. The same weighting formula (based on positive seeds, negative seeds, and entity count) is universally applied to all web-lists regardless of domain. This universal approach achieves precision improvement through consistent quality filtering while avoiding the architectural complexity of maintaining multiple specialized wrappers for different domains.
Data Source
AI summary
A set expansion system is described herein that improves precision, recall, and performance of prior set expansion methods for large sets of data. The system maintains high precision and recall by 1) identifying the qualify of particular lists and applying that quality through a weight, 2) allowing for the specification or negative examples in a set of seeds to reduce the introduction of bad entities into the set, and 3) applying a cutoff to eliminate lists that include a low number of positive matches. The system may perform multiple passes to first generate a good candidate result set and then refine the set to find a set with highest quality. The system may also apply Map Reduce or other distributed processing techniques to allow calculation in parallel. Thus, the system efficiently expands large concept sets from a potentially small set of initial seeds from readily available web data.