Extract Entities from EU Legal Text

This is a set of tools created to extract author names and affiliations from unstructured text data (e.g., first pages of legal documents or academic papers).

Basic Usage Description

👤 EXTRACT Author Names

The first tool, run_extract_authors_basic(), expects a text column (first_page_text) that may contain author names. For each row, the function generates a user message via the default prompt function and sends it to the GPT client. Results are returned as a list of comma-separated author name strings.

from gptquery.tools.tool_text_extraction import run_extract_authors_basic
df = ...  # DataFrame with 'first_page_text'
df_out = run_extract_authors_basic(df, api_key="your-openai-key")
print(df_out['author_names'])
# [["Katarzyna Zieleskiewicz", "Bartlomiej Kurcz"], [], ["ERROR"]]

🏫 EXTRACT Author Affiliations

The second tool, run_extract_affiliations_basic(), also expects a first_page_text column. For each row, it extracts institution and organization names. Results are returned as a list of semicolon-delimited affiliation strings.

from gptquery.tools.tool_text_extraction import run_extract_affiliations_basic
df = ...  # DataFrame with 'first_page_text'
df_out = run_extract_affiliations_basic(df, api_key="your-openai-key")
print(df_out['affiliations'])
# [["University of Groningen", "Court of Justice of the EU"], [], ["ERROR"]]

📤 Input/Output Schema

Input Columns:

Column	Type	Description
first_page_text	str	Raw text from the first page of a document, expected to contain author names and/or affiliations

Output Columns:

• author_names → list of author name strings (comma-split); [] if none found; ["ERROR"] on failure
• affiliations → list of institution/organization strings (semicolon-split); [] if none found; ["ERROR"] on failure

💾 Example DataFrame

doc_id	first_page_text	author_names	affiliations
DOC_001	"Written observations submitted by Katarzyna Zieleskiewicz, University of Groningen..."	["Katarzyna Zieleskiewicz"]	["University of Groningen"]
DOC_002	"Submitted on behalf of the Commission by J. Hottiaux, acting as agent..."	["J. Hottiaux"]	["European Commission"]
DOC_003	(cover page only, no author block)	[]	[]