assembly-bills

60,925 member-proposed bills from the 20th-22nd Korean National Assembly (2016-2026), with full propose-reason texts.

20th Assembly   21,594 bills   2016-05 ~ 2020-05
21st Assembly   23,655 bills   2020-06 ~ 2024-05
22nd Assembly   15,676 bills   2024-06 ~ 2026-02

The propose-reason text (제안이유) was scraped in full from the National Assembly Legislative Information System. 99.4% of bills have text available. This is the first publicly available dataset containing the full propose-reason texts for Korean legislative bills.

Quickstart

Option A: CLI

pip install korean-assembly-bills

If pip doesn't work, try pip3 install korean-assembly-bills or python3 -m pip install korean-assembly-bills. This is common on macOS where pip may not be in your PATH.

assembly-bills info                          # dataset summary
assembly-bills search "인공지능"              # search by keyword
assembly-bills search "부동산" --age 21       # filter by assembly
assembly-bills show 2124567                   # view a bill + full text
assembly-bills mp "이재명"                    # look up an MP's bills
assembly-bills stats --by party              # statistics
assembly-bills export ai.csv --keyword "인공지능" --with-text

Option B: Python

import pandas as pd

bills = pd.read_parquet("data/bills.parquet")
texts = pd.read_parquet("data/bill_texts.parquet")

# merge metadata and text
df = bills.merge(texts[["BILL_ID", "propose_reason"]], on="BILL_ID", how="left")
print(df.shape)  # (60925, 25)

Option C: Interactive App

kr-bills.streamlit.app - browse, search, and explore bills in your browser.

Or run locally:

pip install korean-assembly-bills[app]
streamlit run app.py

Data Format

All files are in Apache Parquet format (compressed with zstd). Parquet is a columnar format that is 5-30x smaller than CSV, faster to load, and preserves data types. It is natively supported by pandas, polars, R (arrow), and most data tools.

Reading Parquet files

Python (pandas)

import pandas as pd
df = pd.read_parquet("data/bills.parquet")

Python (polars)

import polars as pl
df = pl.read_parquet("data/bills.parquet")

R

library(arrow)
df <- read_parquet("data/bills.parquet")

DuckDB (SQL)

SELECT * FROM read_parquet('data/bills.parquet') LIMIT 10;

If you need CSV, use the CLI export command:

assembly-bills export all_bills.csv
assembly-bills export all_with_text.csv --with-text

Data Dictionary

bills.parquet

Bill metadata from the Open National Assembly API. 60,925 rows, 24 columns.

bill_texts.parquet

Propose-reason texts scraped from the legislative information system. 60,925 rows, 3 columns.

proposers.parquet

Individual proposer records linking MPs to bills. 769,773 rows, 14 columns.

mp_metadata.parquet

MP biographical information. 947 rows (661 unique MPs across 20th-22nd assemblies), 16 columns. Sourced from the ALLNAMEMBER API endpoint with per-assembly party data from BILLINFOPPSR.

Note on CMIT_NM: The ALLNAMEMBER API only returns committee assignments for the current (22nd) assembly, regardless of the UNIT_CD parameter. Committee data for the 20th and 21st assemblies is therefore set to blank rather than showing incorrect 22nd-assembly committee names.

CLI Reference

assembly-bills [COMMAND] [OPTIONS]

search options

-n, --limit N       Max results (default 20)
--age N             Filter by assembly (20, 21, 22)
--committee TEXT    Filter by committee (substring)
--party TEXT        Filter by party (substring)
--text              Also search in propose-reason texts

export options

--keyword TEXT      Filter by keyword
--age N             Filter by assembly
--with-text         Include propose-reason text
--format [csv|parquet]  Output format (default csv)

Data Collection

The data was collected from two sources:

1. Open National Assembly API (open.assembly.go.kr): Bill metadata, proposer lists, and MP information. APIs used:

   • nzmimeepazxkubdpn (bill list)
   • BILLINFOPPSR (proposer list)
   • nwvrqwxyaytdsfvhu (MP metadata)

2. Legislative Information System (likms.assembly.go.kr): Propose-reason texts were scraped from bill detail pages. The scraping process:

   • GET the bill detail page to obtain a CSRF token
   • POST with the full form parameters to retrieve the <pre id="prntSummary"> content
   • This two-step process is necessary because the API does not provide the propose-reason text

All 60,925 bills across the 20th-22nd assemblies were scraped, with a 99.4% success rate (60,546 texts obtained).

Known Limitations

• Proposer list cap: The Open Assembly API returns a maximum of 100 proposer records per bill. 208 bills (mostly mass-signature bills with 100+ co-sponsors) have truncated proposer lists. These bills can be identified by having exactly 100 rows in proposers.parquet.
• Committee data for past assemblies: CMIT_NM in mp_metadata.parquet is only available for the 22nd Assembly. The API does not return historical committee assignments for the 20th or 21st Assembly (see note in data dictionary).
• 177 joint lead proposals: Some 22nd Assembly bills have multiple lead proposers (comma-separated RST_MONA_CD). Analysis assuming a single lead proposer per bill should account for these cases.

Use Cases

• Legislative text analysis: NLP on Korean legislative language, topic modeling, text similarity
• Political science research: MP productivity, party differences, committee dynamics
• AI impact studies: Compare legislative texts before and after ChatGPT (Nov 2022)
• Network analysis: Co-sponsorship networks from the proposers table

Citation

If you use this dataset in academic work, please cite:

@misc{yang2026assemblybills,
  author = {Yang, Kyusik},
  title = {Korean National Assembly Bills Dataset (20th-22nd Assembly)},
  year = {2026},
  publisher = {GitHub},
  url = {https://github.com/kyusik-yang/korean-assembly-bills}
}

License

CC BY 4.0. The underlying legislative data is public information from the Korean National Assembly.

Related Projects

• open-assembly-mcp -- MCP server for Korean National Assembly data
• assembly-explorer -- Interactive explorer for Korean legislative data

_{Built with Claude Code. To study bills, you need API bills.}