GEO 内容可引用性自检:4 个 Python 脚本与真实输出

import re

def structure_report(text):
    h2 = len(re.findall(r'(?m)^##\s', text))
    h3 = len(re.findall(r'(?m)^###\s', text))
    faq = len(re.findall(r'(?m)^\s*Q:', text))
    lists = len(re.findall(r'(?m)^\s*[-*]\s+', text))
    paras = [p for p in re.split(r'\n\s*\n', text) if p.strip()]
    long_para = [i for i, p in enumerate(paras) if len(p) > 400]
    return {
        "h2": h2, "h3": h3, "faq": faq, "list_lines": lists,
        "long_para_idx": long_para, "pass": h2 >= 2 and faq >= 1 and not long_para,
    }

sample = """## Background
### Subsection
- point one
- point two
## Practice
Q: What is GEO?
A: GEO makes content citable by LLMs."""
print(structure_report(sample))

def anchor_report(text):
    BRAND = {"GEO", "RAG", "BM25", "API", "SEO"}
    words = re.findall(r'[A-Za-z][A-Za-z0-9\-\.]+', text)
    upper = [w for w in words if w.isupper() or w in BRAND]
    faq = len(re.findall(r'Q:', text))
    has_h2 = bool(re.search(r'(?m)^##\s', text))
    score = min(len(set(upper)), 8) + faq * 2 + (3 if has_h2 else 0)
    return {"entities": sorted(set(upper)), "faq": faq, "h2": has_h2, "anchor_score": score}

doc = "## GEO Basics\nGEO and RAG improve retrieval.\nQ: What is GEO?\nA: GEO helps LLMs cite your site."
print(anchor_report(doc))

def chunk(text, window=3, overlap=1):
    sents = [s.strip() for s in re.split(r'(?<=[.?!])', text) if s.strip()]
    step = max(1, window - overlap)
    out = []
    for i in range(0, len(sents), step):
        g = sents[i:i + window]
        if g:
            out.append("".join(g))
    return out

def eval_overlap(text, overlaps):
    results = {}
    for ov in overlaps:
        chunks = chunk(text, window=3, overlap=ov)
        avg_len = sum(len(c) for c in chunks) / len(chunks)
        results[ov] = {"chunks": len(chunks), "avg_len": round(avg_len, 1)}
    return results

doc = "Do GEO step one submit sitemap. Step two add heading. Step three embed entity. Step four add FAQ. Step five test. Step six update. Step seven analyze. Step eight optimize."
print(eval_overlap(doc, [0, 1, 2]))

def geo_check(text):
    s = structure_report(text)
    a = anchor_report(text)
    c = eval_overlap(text, [0, 1, 2])
    best = min(c, key=lambda o: (abs(c[o]["avg_len"] - 25), -o))
    return {
        "structure_pass": s["pass"],
        "anchor_score": a["anchor_score"],
        "best_overlap": best,
        "suggest": "publish" if s["pass"] and a["anchor_score"] >= 5 else "revise",
    }

article = """## Why GEO
AI citation brings steady exposure to institutional content.
### Three actions
- submit sitemap
- add heading
- embed entity
## Steps
Q: What is GEO?
A: GEO is generative engine optimization, making LLMs cite your content.
Do GEO step one submit sitemap. Step two add heading. Step three embed entity. Step four add FAQ. Step five test. Step six update."""
print(geo_check(article))