# RAIDP03-BP05 Review the correctness of the content of your datasets
<a name="raidp03-bp05"></a>

 Create regular review processes for ground-truth labels and factual content across your datasets. Implement fact-checking procedures using human reviewers or comparison against authoritative sources to identify and correct inaccuracies. Datasets used for veracity evaluation may require high accuracy standards to provide reliable measurements. Document the review process and track accuracy metrics over time, updating datasets when new information becomes available or when errors are discovered. 

 **Level of risk exposed if this best practice is not established:** High 

## Implementation considerations
<a name="implementation-considerations-55"></a>

1.  Design your datasets with built-in accuracy validation by enabling multiple sources to confirm factual claims before including them. 

1.  Create fact-checking workflows that combine domain experts with authoritative source verification during dataset creation. Have subject matter experts review content and flag potential inaccuracies before data gets finalized. 

1.  Apply stricter standards to datasets that will be used for evaluation, since these provide the ground truth for measuring release criteria. Engage multiple reviewers to validate each claim and achieve high agreement before accepting labels. 

1.  Schedule periodic reviews of your dataset content to catch errors that may have emerged over time or due to changing information. Plan regular audits where you re-examine your data to verify labels and factual claims are still accurate. 

1.  Build correction processes for when you discover errors or when new information becomes available that affects your dataset accuracy. Create clear workflows for updating factual content and maintaining dataset integrity over time. 

## Resources
<a name="resources-52"></a>

 **Related documents:** 
+  [Visualize data quality scores and metrics generated by AWS Glue Data Quality](https://aws.amazon.com/blogs/big-data/visualize-data-quality-scores-and-metrics-generated-by-aws-glue-data-quality/) 
+  [Build a data quality score card using AWS Glue DataBrew, Amazon Athena, and Quick](https://aws.amazon.com/blogs/big-data/build-a-data-quality-score-card-using-aws-glue-databrew-amazon-athena-and-amazon-quicksight/) 
+  [Ground truth generation and review best practices for evaluating generative AI question-answering with FMEval](https://aws.amazon.com/blogs/machine-learning/ground-truth-generation-and-review-best-practices-for-evaluating-generative-ai-question-answering-with-fmeval/) 
+  [Inspect your data labels with a visual, no code tool to create high-quality training datasets with Amazon SageMaker Ground Truth Plus](https://aws.amazon.com/blogs/machine-learning/inspect-your-data-labels-with-a-visual-no-code-tool-to-create-high-quality-training-datasets-with-amazon-sagemaker-ground-truth-plus/) 
+  [ISO/IEC 42001:2023](https://www.iso.org/standard/42001)A.7.2 Data for development and enhancement of AI system 
+  [ISO/IEC 42001:2023](https://www.iso.org/standard/42001) A.7.4 Quality of data for AI systems 

 **Related tools:** 
+  [Amazon Bedrock Guardrails : Use contextual grounding check to filter hallucinations in responses](https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails-contextual-grounding-check.html) 
+  [The Effects of Data Quality on Machine Learning Performance on Tabular Data](https://arxiv.org/abs/2207.14529) 
+  [A Survey on Data Quality Dimensions and Tools for Machine Learning](https://arxiv.org/abs/2406.19614) 
+  [BoundingDocs: a Unified Dataset for Document Question Answering with Spatial Annotations](https://arxiv.org/pdf/2501.03403v1) 
+  [CodeUltraFeedback: An LLM-as-a-Judge Dataset for Aligning Large Language Models to Coding Preferences](https://arxiv.org/pdf/2403.09032v3)