Confidence score
Higher confidence scores generally correspond to a heightened level of certainty and accuracy. Conversely, lower scores may indicate a greater degree of uncertainty or potential errors in the data. Confidence scores range from zero (indicating no confidence) to one (representing very high confidence). The confidence_scores
feature empowers you to make informed decisions regarding Ocrolus' Instant response. You have the option to entirely reject Instant results and instead choose the Complete mode, based on your assessment of the extraction's quality.
After the fields are extracted from a document using Forms and Classification workflow endpoints, Ocrolus provides confidence scores for each extracted field. These confidence_scores
represent a set of values or scores that indicate the level of certainty or confidence associated with the extracted and processed data presented in the response. These scores are commonly used to convey the reliability or accuracy of the information provided.
Key benefits
Here are several typical business scenarios that you can address using the confidence_scores
feature:
- Optimized data quality: By utilizing confidence scores, you can confidently eliminate extractions with low confidence levels. This enhances the accuracy of your extracted data, improving the quality of insights and decision-making.
- Efficient human review process: Insert medium-confidence extractions into interfaces for human review. This accelerates the review process by focusing efforts on fields that require attention, streamlining the overall workflow.
- Data confidence-based decision-making: Leverage confidence scores to select the most reliable extracted value from a group of interchangeable fields. This empowers you to make informed decisions with the utmost certainty.
- Enhanced response evaluation: With confidence scores, you have the flexibility to reject Instant responses in favor of the Complete mode. This ensures optimal accuracy by directing applications toward comprehensive human review when needed.
Updated about 1 year ago