Medical Term Ambiguity Scoring via Conditional Probability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated hyperlink technologies, such as those used in the Westlaw legal research system, are not effective in identifying medical terms due to their ambiguity across different contexts, unlike legal citations and names which function consistently.
Innovation Solution
A system and method utilizing a term-ambiguity calculator with a scoring function based on conditional probabilities from language models to determine whether a term is medical or non-medical, incorporating the Unified Medical Language System (UMLS) and applying ngram backoff with Witten Bell smoothing to calculate ambiguity scores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If automated hyperlink technology is applied to identify medical terms, then document linking capability is improved, but identification accuracy deteriorates due to term ambiguity
Solution Approach 1:
The patent introduces an ambiguity score as an intermediary metric to bridge the gap between automated term identification and accurate medical term recognition. The ambiguity score quantifies the uncertainty of each term's classification, enabling the system to distinguish between clear-cut medical terms and ambiguous terms that require contextual analysis. This intermediary measure allows automated processing while maintaining accuracy by flagging uncertain cases for further review or applying context-aware disambiguation rules.
2Speed
If context-independent term identification is used, then processing speed is improved, but term classification accuracy deteriorates
Solution Approach 1:
The patent implements preliminary action by pre-calculating and storing ambiguity scores for terms in a database before actual document processing. During runtime, the system quickly retrieves pre-computed ambiguity scores rather than performing complex contextual analysis for each term. This preliminary preparation enables fast processing while maintaining accuracy, as the pre-analyzed ambiguity information captures contextual nuances without requiring real-time computational overhead.
3Device complexity
If simple term matching is applied, then system complexity is reduced, but identification reliability deteriorates for ambiguous medical terms
Solution Approach 1:
The patent applies parameter changes by transforming the binary classification problem (medical vs. non-medical) into a continuous ambiguity score spectrum. Instead of simple yes/no matching, the system evaluates terms on a scale of ambiguity, allowing nuanced differentiation between confidently identified medical terms and ambiguous ones. This parameter transformation maintains computational simplicity while dramatically improving reliability by providing graded confidence levels that reflect the true uncertainty in medical term identification.
Data Source
AI summary
Some known medical terms may function as non-medical terms depending on their particular context. Accordingly, the present inventors devised systems, methods, and software that facilitate determining whether a term that is found in a medical corpus is likely to be a medical term when found in another corpus. An exemplary embodiment receives a term and computes an ambiguity score based on language models for a medical and a non-medical corpus.