Line-Item Consolidation for Form 1120

Overview

Development stage

DeployedSource: c) Deployed – The use case is being actively authorized or utilized to support the functions or mission of an agency.

High-impact designation

Not high-impactSource: c) Not high-impact

Impact justification

The output is not presumed to be high-impact and is not used as the principal basis for significant decisions/actions

Start date

2024-09-27 (day precision)Source: 9/27/2024

Withheld from public reporting

NoSource: a) No

Mission

Topic area

Administrative Functions

Operational functions

Document processing and extraction Derived by keyword rules; see methodology

Problem the AI is intended to solve

Large Business & International (LB&I) partnerships have a variety of forms and statements that include text. Filings include semi-structured attachments to returns that consist of many short free text labels paired with dollar amounts. Short text labels that include information like descriptions, addresses, and alphanumeric type codes. Non-standard labels are difficult to structure and interpret unstructured narrative text, ranging from a sentence to entire paragraph-based documents, that taxpayers provide to explain, justify, or document a position. This is a research project. The goal of this project is to identify opportunities to derive value from the unstructured text. Assess potential value to examiners from consolidation across labeled pairs, utilize consolidated line items for outlier analysis, etc.

Expected benefits

Line-item consolidation is a crucial step in providing value to examiners through the grouping of numerous free-text responses into common, recognizable groups that allow for further analysis and exploration. Not only do these line-items become more accessible to examiners, but it facilitates a more robust benchmarking analysis of line-item amounts by creating larger groupings of common line-items. Success in the line-item consolidation stage was defined by how effectively each stage of the text consolidation pipeline could reduce the number of unique line-item descriptions while maintaining an appropriate level of specificity.

System outputs

A Form 1120 Other Deduction line-item description is assigned or consolidated to a consolidation group, which is either a subject matter expert defined deduction category or a deduction description existing in the dataset, based on how semantically similar the descriptions are, measured by cosine similarity. The model produces an additional column appended to the original dataset that contains the consolidation group (text string) that the line-item description (row) was matched to.

Technology

AI classification

Natural language processingSource: Natural Language Processing: AI that processes, interprets, and shares information in human language.

System name(s)

Not reported

Custom-developed code

Yes

Public source code

Not reported

Sourcing

How it was built

Developed in-houseSource: b) Developed in-house

Vendor (as reported)

Not reported

Vendors (standardized)

None on the standard list

A vendor is the supplier named by the agency. It does not identify the underlying model or AI technology.

Data and privacy

Involves PII

No

Privacy Impact Assessment

Not reported

Authorization to Operate

Yes

Demographic features

k) None of the above

Training and evaluation data

The team evaluated the appropriateness of assignments in the top 500 most frequent and material groups in order to further quantify the accuracy of consolidation. Original descriptions were manually reviewed within their respective assigned consolidated groups and a judgment was made on whether the description fit within the rest of the group or not. We also evaluated the amount of consolidation that our updated algorithm could perform. On a test of all 1120 Other Deductions in CDW, the algorithm was able to consolidate nearly 4 million unique line-item descriptions into 253 groups.

Federal Data Catalog

Not reported

Governance

0 of 8 minimum-practice questions answered. A blank answer means the agency reported nothing; it does not mean the practice is absent.

Potential impacts and how they were identified

Not reported