Result Set Options

Can be one of the following types:

The 'Value Type' determines how extracted values are parsed and formatted.

• Use this to enforce that results conform to a specific type (such as date, number, or boolean).
• Values that cannot be converted to the specified type, or that fall outside a valid range, can be excluded from the result set.
• When a formattable type is selected, output values are normalized to a standard representation (e.g., dates as YYYY-MM-DD, numbers with thousands separators).
• Sorting by value is also enabled for numeric or date types.

Example:
To ensure only valid dates are returned, set 'Value Type' to a date type. Invalid values (such as 99-99-9999) will be filtered out.

Use 'Value Override' to unify multiple variations of a value to a single, consistent output.

• This is useful for classification features, where you want to map different document titles or entity names to a single feature value.
• Also helpful in data extraction to standardize company names, product codes, or other fields with known variations.

Example:
Map "Change of Address", "Address Change Form", and "Notice of Address Change" to "feature_coa_title".

A combination of the following flags:

Set 'Case Normalization' to enforce a specific casing (such as uppercase, lowercase, or title case) on all output values.

• Useful for standardizing data for export, comparison, or downstream processing.
• If set to None, no case transformation is applied.

Example:
Convert "Acme Roofing, Inc." to "ACME ROOFING, INC." by selecting uppercase.

Can be one of the following values:

'Word Normalization' can pluralize, singularize, or stem words in the output value.

• Use to reduce inflected forms to a common base, improving consistency for classification or search.
• Porter stemming is English-only and reduces words to their root form (e.g., "reports", "reported", "reporting" → "report").
• Pluralize or singularize to enforce a specific grammatical number.

Note:
Word normalization is only applied when the value consists of one or more words separated by whitespace.

Can be one of the following types:

The 'Subtractor' allows you to define an extractor that removes unwanted content from each output value.

• The extractor is run against each value after all other normalization steps.
• Any matching content is removed; if the result is empty or whitespace, the value is discarded.
• The extractor must match contiguous sequences of characters.

Example:
Remove trailing punctuation or formatting codes from extracted values.

Can be one of the following types:

When enabled, the region associated with each result is adjusted to align with detected lines (such as table or form boundaries).

• Depends on Layout Data generated by Line Detection or Line Removal during image processing or recognition.
• Use to improve the accuracy of region-based extraction, especially for structured documents.
• See Result Snap Options for additional configuration.

Use 'Confidence Override' to set a fixed confidence score for all results, regardless of their original value.

• Useful for boosting or reducing the weighting of specific features in classification or extraction.
• Set to a value greater than 0 to apply; otherwise, the original confidence is used.

Example:
Set to 0.8 to assign 80% confidence to all results.

The 'Confidence Multiplier' is applied after any override, scaling the final confidence score.

• Use to boost or reduce the influence of results in downstream processing.
• Set to 1 for no change; values greater than 1 increase confidence, values less than 1 decrease it.

Example:
Set to 0.5 to halve all confidence scores.

Use Case: Lexical Classification

In lexical classification, certain features—such as a document title like "CHANGE OF ADDRESS" centered at the top of a form—can be strong indicators of a specific document type (for example, a change of address form). When an extractor feeds features into the classification engine, you may want to boost the results from certain extractors to reflect their importance.

The 'Confidence Multiplier' property allows you to easily apply such a boost. For example, to make the presence of a "CHANGE OF ADDRESS" title five times more influential in classification, set this property to 5.

This approach helps ensure that highly indicative features have a proportionally greater impact on classification outcomes.

Use Case: Prioritizing Extractors in a Data Type

When a Data Type is configured with multiple extractors and outputs results ordered by confidence, you can use 'Confidence Multiplier' to prefer the results of specific extractors over others.

• By assigning a higher multiplier to a preferred extractor, its results will have higher confidence and appear at the top of the result set when sorted by confidence.
• This enables designers to control which extraction logic is favored, making it easy to implement fallback or tiered extraction strategies.

Example:
If you have a primary extractor for a field and a secondary fallback extractor, set the primary's multiplier to 2 and the fallback's to 1. Results from the primary extractor will be preferred in the output.

A combination of the following flags:

'Processing Options' control string cleanup, group flattening, and whether to consider character-level OCR confidence.

• Options include trimming whitespace, cleaning control characters, flattening groups, and more.
• Combine multiple options as needed to achieve the desired output.

Use 'Hit Count Range' to enforce that the number of results falls within a specified range.

• If the result count is outside the range, the entire result set is discarded.
• Useful for validation scenarios where a certain number of hits is required.

Example:
Require at least 1 and no more than 3 results by setting the range to 1-3.

Use 'Sort Order' to define how results are ordered in the output.

• Sort by value, position, confidence, or other criteria.
• Multiple sort specifications can be combined for complex ordering.
• Sorting is applied after value normalization and filtering.

Example:
Sort by date descending, then by confidence.

When 'Distinct' is enabled, only the first occurrence of each unique value is retained.

• Comparison is case-insensitive and applied after all normalization steps.
• Use to ensure that each value appears only once in the output.

Use 'Truncate At' to restrict the result set to a specific number of items.

• Set to 0 to return all results.
• If the result set exceeds this number, only the first N results (after sorting and filtering) are returned.

Example:
Set to 1 to return only the top result.