parse_ind

topic_segmentation.news_dataset.parse_ind

Parse saved pages in data/pages/independent_subheadings/ into data/outputs/ind_segments.json.

is_noise_heading(text: str) -> bool

Source code in src/topic_segmentation/news_dataset/parse_ind.py

def is_noise_heading(text: str) -> bool:
    t = text.lower().strip()
    return len(t) < 3 or any(pat in t for pat in NOISE_PATTERNS)

is_skip_container(el) -> bool

Source code in src/topic_segmentation/news_dataset/parse_ind.py

def is_skip_container(el) -> bool:
    if el.name in ("aside", "nav", "form"):
        return True
    cls = set(el.get("class") or [])
    if cls & SKIP_DIV_CLASSES:
        return True
    return False

is_link_only_list(el) -> bool

Source code in src/topic_segmentation/news_dataset/parse_ind.py

def is_link_only_list(el) -> bool:
    if el.name not in ("ul", "ol"):
        return False
    items = el.find_all("li", recursive=False)
    if not items:
        return False
    for li in items:
        li_text = li.get_text(" ", strip=True)
        a_text = "".join(a.get_text(" ", strip=True) for a in li.find_all("a"))
        if not li_text or not a_text:
            return False
        if li_text == a_text:
            continue
        if len(a_text) / len(li_text) >= 0.6:
            continue
        return False
    return True

is_trailing_element(el) -> bool

Source code in src/topic_segmentation/news_dataset/parse_ind.py

def is_trailing_element(el) -> bool:
    if el.name in ("h2", "h3", "h4"):
        t = el.get_text(" ", strip=True).lower().rstrip(":")
        if t in TRAILING_HEADINGS:
            return True
    text_lower = el.get_text(" ", strip=True).lower()
    if any(text_lower.startswith(p) for p in TRAILING_TEXT_PREFIXES):
        return True
    return False

clean_inner_html(el) -> str

Source code in src/topic_segmentation/news_dataset/parse_ind.py

def clean_inner_html(el) -> str:
    el_copy = copy.deepcopy(el)
    for tag in el_copy.find_all(True):
        if tag.name == "script":
            tag.decompose()
            continue
        for attr in STRIP_ATTRS:
            tag.attrs.pop(attr, None)
        if tag.name == "a" and (tag.get("href") or "").startswith("/"):
            tag["href"] = "https://www.independent.co.uk" + tag["href"]
            tag["target"] = "_blank"
            tag["rel"] = "noopener"
    return el_copy.decode_contents().strip()

collect_elements(container) -> list

Collect content elements from the article body.

Source code in src/topic_segmentation/news_dataset/parse_ind.py

def collect_elements(container) -> list:
    """Collect content elements from the article body."""
    elements = []
    for el in container.children:
        if not hasattr(el, "name") or el.name is None:
            continue
        if is_skip_container(el):
            continue

        cls = set(el.get("class") or [])

        # Section subheading div or content list wrapped in sc-kk992l-0
        if el.name == "div" and "sc-kk992l-0" in cls:
            inner_list = el.find(["ul", "ol"])
            inner_head = el.find(["h2", "h3", "h4"])
            if inner_list and not inner_head:
                # Include content lists and skip link-only "Read more" lists.
                if not is_link_only_list(inner_list):
                    elements.append(inner_list)
            else:
                elements.append(el)
            continue

        # Image div (contains a figure with an actual image)
        if el.name == "div" and "image" in cls:
            fig = el.find("figure")
            if fig:
                img = fig.find("img")
                if img and (img.get("src") or "").startswith("http"):
                    elements.append(fig)
            continue

        # Content elements
        if el.name in ("p", "ul", "ol", "blockquote"):
            if not is_link_only_list(el):
                elements.append(el)
            continue

        # Skip unrecognized divs, including ads and empty wrappers.

    return elements

process_article(aid: str, raw_html: str, meta: dict)

Source code in src/topic_segmentation/news_dataset/parse_ind.py

def process_article(aid: str, raw_html: str, meta: dict):
    soup = BeautifulSoup(ftfy.fix_text(raw_html), "html.parser")

    article = soup.find("article")
    if not article:
        return None

    # Article header (inside article, contains title/standfirst/hero)
    art_header = article.find("header")

    # Title
    h1 = art_header.find("h1") if art_header else article.find("h1")
    title = h1.get_text(" ", strip=True) if h1 else meta.get("title", "")

    # Standfirst: h2.sc-1cmj4ml-0 inside article header
    standfirst_text = None
    if art_header:
        sf = art_header.find("h2", class_=lambda c: c and "sc-1cmj4ml-0" in c)
        if sf:
            text = sf.get_text(" ", strip=True)
            if text and not text.lower().startswith("sign up"):
                standfirst_text = text

    # Hero image from article header figure
    hero_img_src = None
    hero_img_cap = None
    if art_header:
        fig = art_header.find("figure")
        if fig:
            img = fig.find("img")
            if img and (img.get("src") or "").startswith("http"):
                hero_img_src = img["src"]
                figcap = fig.find("figcaption")
                if figcap:
                    hero_img_cap = figcap.get_text(" ", strip=True)

    # Content div
    main_wrapper = article.find("div", class_=lambda c: c and "main-wrapper" in c)
    if not main_wrapper:
        return None
    content = main_wrapper.find("div", class_=lambda c: c and "sc-jbiisr-0" in c)
    if not content:
        return None

    sentences, html_parts, labels, title_labels = [], [], [], []

    if standfirst_text:
        sentences.append(standfirst_text)
        html_parts.append({"tag": "p", "html": standfirst_text})
        labels.append(0)
        title_labels.append(1)

    if hero_img_src:
        cap_html = f"<figcaption>{hero_img_cap}</figcaption>" if hero_img_cap else ""
        sentences.append(hero_img_cap or "")
        html_parts.append({"tag": "figure", "html": f'<img src="{hero_img_src}" style="max-width:100%">{cap_html}'})
        labels.append(0)
        title_labels.append(0)

    first_elem = not (standfirst_text or hero_img_src)

    for el in collect_elements(content):
        if is_trailing_element(el):
            break

        # Figure (body image)
        if el.name == "figure":
            img = el.find("img")
            if not img:
                continue
            src = img.get("src", "")
            if not src:
                continue
            figcap = el.find("figcaption")
            cap_text = figcap.get_text(" ", strip=True) if figcap else ""
            cap_html = f"<figcaption>{cap_text}</figcaption>" if cap_text else ""
            sentences.append(cap_text)
            html_parts.append({"tag": "figure", "html": f'<img src="{src}" style="max-width:100%">{cap_html}'})
            labels.append(0)
            title_labels.append(0)
            first_elem = False
            continue

        # Section subheading div (sc-kk992l-0)
        if el.name == "div" and "sc-kk992l-0" in set(el.get("class") or []):
            text = el.get_text(" ", strip=True)
            if not text or is_noise_heading(text):
                continue
            if len(text) > 200:
                continue
            inner_head = el.find(["h2", "h3", "h4"])
            tag = inner_head.name if inner_head else "h3"
            sentences.append(text)
            html_parts.append({"tag": tag, "html": text})
            labels.append(0 if first_elem else 1)
            title_labels.append(1)
            first_elem = False
            continue

        text = el.get_text(" ", strip=True)
        if not text:
            continue

        if any(text.lower().startswith(p) for p in SKIP_PARA_PREFIXES):
            continue

        sentences.append(text)
        html_parts.append({"tag": el.name, "html": clean_inner_html(el)})
        labels.append(0)
        title_labels.append(0)
        first_elem = False

    # Remove trailing headings without following content.
    while sentences and title_labels[-1] == 1:
        sentences.pop()
        html_parts.pop()
        labels.pop()
        title_labels.pop()

    if not sentences:
        return None

    return {
        "articleId": aid,
        "outlet": "Independent",
        "title": title,
        "url": meta.get("url", ""),
        "sentences": sentences,
        "html_parts": html_parts,
        "labels": labels,
        "title_labels": title_labels,
        "num_sections": labels.count(1) + 1,
    }

main()

Source code in src/topic_segmentation/news_dataset/parse_ind.py

def main():
    with open(ROOT / "data/articles.json", encoding="latin-1") as f:
        all_meta = {a["articleId"]: a for a in json.load(f)}

    results = []
    seen_content: set = set()
    skipped = 0
    for html_file in sorted((ROOT / "data/pages/independent_subheadings").glob("*.html")):
        aid = html_file.stem
        raw = html_file.read_text(encoding="utf-8")
        rec = process_article(aid, raw, all_meta.get(aid, {}))
        if not rec:
            skipped += 1
            continue
        content = (rec.get("title", "") + "".join(rec["sentences"]))
        if content in seen_content:
            skipped += 1
            continue
        seen_content.add(content)
        results.append(rec)

    one_section = sum(1 for r in results if r["num_sections"] == 1)
    results = [r for r in results if r["num_sections"] >= 2]
    print(f"Processed {len(results)} articles ({skipped} skipped/duplicate)")
    print(f"  1 section: {one_section}, 2+ sections: {len(results) - one_section}")

    (ROOT / "data/outputs").mkdir(parents=True, exist_ok=True)
    seg_path = ROOT / "data/outputs/ind_segments.json"
    with open(seg_path, "w", encoding="utf-8") as f:
        json.dump(results, f, ensure_ascii=False, indent=2)
    print(f"Saved {seg_path}")