Confidence score
A confidence score assesses a model prediction according to a particular method. It does not grant business approval.
Understand what a score assesses
A confidence score describes how certain a model considers a specific prediction according to its method. Its scale and meaning must be established for the actual system. In document recognition, for example, scores for a word and an assigned field may differ.
In a fictional example, the price 24 EUR is recognised with a field score of 0.98. That assesses the relevant recognition. It does not confirm that 24 EUR matches the valid contract or that the whole invoice is correct.
A threshold can route results for review. Its suitability must be examined using relevant examples and the consequences of errors. The score is also not a statistical confidence interval or a general guarantee of a particular error rate in your business.
Sources
Related terms
- Optical character recognitionOptical character recognition makes writing in images machine-processable. Reading an amount is not the same as checking an invoice against business rules.
- Re-rankingRe-ranking reassesses search candidates already retrieved. Relevance to the query and the limits of the initial selection both matter.
- Document chunkingSplitting a document into smaller coherent sections for search and model context.
- Hybrid searchCombining lexical search and vector search into a shared result list.
- Semantic searchSearch that finds and ranks results by their semantic relevance to the query.
- Model routingChoosing a suitable model for each request from a defined set of permitted candidates.
Articles on this topic
Your project with ASES
Describe your project to us. We will get back to you with an initial assessment.