Multi-Level Passage Ranking for Context-Aware Search Precision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI-based information retrieval models are limited in document size processing and passage-based search methods result in irrelevant document rankings when documents are divided into passages, leading to suboptimal search results.
Innovation Solution
A method integrating document-level and passage-level search results using multiple search models, including a first model for fast processing and high recall, and a second AI-based model for re-ranking, followed by rank fusion and aggregation based on passage and document relationships, to determine a final ranking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If passage-based search method is used to process large documents with AI models, then search precision is improved, but document context is lost leading to irrelevant rankings
Solution Approach 1:
The patent divides documents into passages for AI model processing to improve search precision, while maintaining document-level indexing to preserve context information. This segmentation allows the system to process large documents with token limits while retaining the ability to evaluate overall document relevance.
Solution Approach 2:
The patent implements a nested structure where passages are embedded within their parent documents. The passage retrieval results are mapped back to document levels, allowing the system to leverage both fine-grained passage matching and coarse-grained document context for final ranking.
2Loss of information
If document-based search method is used, then document context is preserved, but search precision decreases for large documents beyond model processing limits
Solution Approach 1:
The patent segments large documents into smaller passages that fit within AI model token limits (e.g., 512 tokens for BERT), enabling precise matching while preserving the ability to evaluate full document relevance through document-level indexing and integrated ranking.
Solution Approach 2:
The patent merges passage-level retrieval results with document-level evaluation by integrating rankings from both levels. This combination allows the system to benefit from precise passage matching while maintaining document context for final relevance assessment.
3Productivity
If only passage-based search is used, then processing speed is improved, but recall decreases due to missing relevant documents
Solution Approach 1:
The patent performs preliminary passage-level retrieval to quickly identify candidate passages, then uses document-level evaluation to ensure comprehensive coverage. This two-stage approach filters results efficiently while maintaining high recall through document-level verification.
Solution Approach 2:
The patent implements feedback loops where passage retrieval results inform document-level evaluation, and document-level results refine passage selection. This iterative process ensures that speed optimizations do not compromise recall by continuously validating results at both levels.
Data Source
AI summary
A method is presented for enhancing search results by segmenting documents into smaller passages and utilizing those passages as the search unit. This method integrates the passage rankings from two search models to produce a new document ranking and arranges the documents accordingly. The ranking of document is also rearranged based on the proportion of passages taken from the same document versus the total number of passages in that document. The final search ranking system combines document-level and passage-level search rankings to rank documents. This method ensures that when conducting a passage search, the search results consider the general relevance of the entire document, which leads to better performance compared to searching only by document-level or passage-level searches.