convert.py

import csv
import os
import re

from datetime import datetime
from glob import glob

from bs4 import BeautifulSoup


BASE_DIR = "reuters21578/"


def clean_date(date_str):
    """
    Date format: 19-OCT-1987 01:51:51.69
    """
    pattern = r"(\d{2}-[A-Z]{3}-\d{4} \d{2}:\d{2}:\d{2}\.\d+)"

    match = re.search(pattern, date_str.strip())

    if match:
        date_str = match.group(1)
        return datetime.strptime(date_str, "%d-%b-%Y %H:%M:%S.%f").isoformat()
    return None


def clean_text(text):
    lines = text.split("\n")
    cleaned_lines = []

    cleaned_lines.append(lines[0])

    for line in lines[1:]:

        # Ignore empty lines
        if not line.strip():
            continue

        if line[0] == " ":
            cleaned_lines.append(line)
        else:
            cleaned_lines[-1] += " " + line

    return "\n\n".join(cleaned_lines[:-1])  # The last line is always "REUTER"


def parse_sgm(fname):
    with open(fname, "r", encoding="ISO-8859-15") as f:
        contents = f.read()

    soup = BeautifulSoup(contents, "html.parser")
    rows_train = []
    rows_test = []

    for meta in soup.find_all("reuters"):
        data = parse_document(meta)

        if data["attr__lewissplit"] == "TRAIN":
            rows_train.append(data)
        if data["attr__lewissplit"] == "TEST":
            rows_test.append(data)

    return rows_train, rows_test


def parse_document(meta):

    # date
    date = meta.find("date").text

    # topics
    topics = [topic.text for topic in meta.find("topics").find_all("d")]

    # places
    places = [place.text for place in meta.find("places").find_all("d")]

    # people
    people = [people.text for people in meta.find("people").find_all("d")]

    # orgs
    orgs = [org.text for org in meta.find("orgs").find_all("d")]

    # exchanges
    exchanges = [exchange.text for exchange in meta.find("exchanges").find_all("d")]

    # companies
    companies = [company.text for company in meta.find("companies").find_all("d")]

    text = meta.find("text")
    text_title = text.find("title")
    text_dateline = text.find("dateline")
    text_body = text.find("body")

    return {
        "attr__topics": meta.attrs["topics"],
        "attr__lewissplit": meta.attrs["lewissplit"],
        "attr__cgisplit": meta.attrs["cgisplit"],
        "attr__oldid": int(meta.attrs["oldid"]),
        "attr__newid": int(meta.attrs["newid"]),
        "date": clean_date(date),
        "topics": topics,
        "places": places,
        "people": people,
        "orgs": orgs,
        "exchanges": exchanges,
        "companies": companies,
        "text__type": text.attrs["type"] if "type" in text.attrs else None,
        "text__title": text_title.text if text_title else None,
        "text__dateline": text_dateline.text if text_dateline else None,
        "text__body": text_body.text if text_body else None,
        "text": clean_text(text_body.text) if text_body else None,
    }


def save_csv(rows, fname):
    """
    Save the processed data into a CSV file.
    """
    with open(fname, "w", encoding="utf8") as f:
        writer = csv.DictWriter(f, fieldnames=rows[0].keys())
        writer.writeheader()

        for row in rows:
            writer.writerow(row)


def run():
    rows_train, rows_test = [], []

    for fname in glob(os.path.join(BASE_DIR, "*.sgm")):
        train, test = parse_sgm(fname)
        rows_train.extend(train)
        rows_test.extend(test)

    save_csv(rows_train, "train.csv")
    save_csv(rows_test, "test.csv")


if __name__ == "__main__":
    run()