unlabelled-data / README.md
enlihhhhh's picture
Update README.md
de60453 verified
|
Raw
History Blame Contribute Delete
1.73 kB
metadata
task_categories:
  - text-generation
  - question-answering
language:
  - en

HTD Scraped Datasets

Cleaned text chunks from Home Team Department (HTD) agency websites, processed into a consistent 8-column format for downstream tasks (QA pairs annotation, fine-tuning, etc.)

Schema

Column Type Description
chunk_text str Cleaned text chunk (40-650 words)
source_url str Original page URL
page_title str Page title
section_headers str Section heading path
heading_path str Hierarchical path from site structure
content_type str guide, press_release, faq, service_index, or event
last_updated str Article date or scrape date
word_count int Word count of chunk_text

Datasets

Agency File Rows Source
HTX (Home Team Science & Technology Agency) htx_dataset_cleaned.parquet 59 htx.gov.sg
MHA (Ministry of Home Affairs) mha_dataset_cleaned.parquet 6,149 mha.gov.sg
SPF (Singapore Police Force) spf_dataset_cleaned.parquet 6,191 police.gov.sg
CNB (Central Narcotics Bureau) cnb_dataset_cleaned_060526.csv 518 cnb.gov.sg
ICA (Immigration & Checkpoints Authority) ica_dataset_cleaned_060526.csv 896 ica.gov.sg
SPS (Singapore Prison Service) sps_filtered_qa_utf8.csv 766 sps.gov.sg

Usage

import pandas as pd

# Load a single dataset
df = pd.read_parquet("htx_dataset_cleaned.parquet")
print(df.head())

Or load via HuggingFace datasets:

from datasets import load_dataset

# Load a specific config
ds = load_dataset("ops-tuned-llm/unlabelled-data")