src/data/normalize_text.py

from __future__ import annotations

import hashlib
import re
import unicodedata
from typing import Iterable


WHITESPACE_RE = re.compile(r"\s+")
PUNCT_RE = re.compile(r"[^\w\s]", flags=re.UNICODE)
SENTENCE_BOUNDARY_RE = re.compile(r"(?<=[.!?。！？])\s+")


def normalize_whitespace(text: object) -> str:
    if text is None:
        return ""
    return WHITESPACE_RE.sub(" ", str(text).replace("\x00", " ")).strip()


def normalize_for_hash(text: object) -> str:
    normalized = unicodedata.normalize("NFKC", normalize_whitespace(text)).casefold()
    normalized = normalized.replace("’", "'").replace("‘", "'")
    normalized = normalized.replace("“", '"').replace("”", '"')
    normalized = normalized.replace("–", "-").replace("—", "-")
    normalized = PUNCT_RE.sub(" ", normalized)
    return normalize_whitespace(normalized)


def stable_hash(text: object, length: int = 16) -> str:
    return hashlib.sha1(normalize_for_hash(text).encode("utf-8")).hexdigest()[:length]


def stable_hash_raw(*parts: object, length: int = 16) -> str:
    raw = "\u241f".join(normalize_whitespace(part) for part in parts)
    return hashlib.sha1(raw.encode("utf-8")).hexdigest()[:length]


def word_count(text: object) -> int:
    text = normalize_whitespace(text)
    if not text:
        return 0
    return len(text.split())


def split_sentences(text: object) -> list[str]:
    text = normalize_whitespace(text)
    if not text:
        return []
    parts = [normalize_whitespace(part) for part in SENTENCE_BOUNDARY_RE.split(text)]
    return [part for part in parts if part]


def split_long_text_by_words(text: str, max_words: int, overlap_words: int = 30) -> list[str]:
    words = text.split()
    if len(words) <= max_words:
        return [text] if text else []
    chunks: list[str] = []
    step = max(1, max_words - overlap_words)
    for start in range(0, len(words), step):
        chunk_words = words[start : start + max_words]
        if chunk_words:
            chunks.append(" ".join(chunk_words))
        if start + max_words >= len(words):
            break
    return chunks


def make_sentence_chunks(
    sentences: Iterable[str],
    max_words: int = 180,
    overlap_sentences: int = 1,
) -> list[dict[str, object]]:
    sentence_list = [normalize_whitespace(sentence) for sentence in sentences if normalize_whitespace(sentence)]
    chunks: list[dict[str, object]] = []
    i = 0
    while i < len(sentence_list):
        current: list[str] = []
        start_i = i
        total_words = 0
        while i < len(sentence_list):
            sentence_words = word_count(sentence_list[i])
            if current and total_words + sentence_words > max_words:
                break
            if not current and sentence_words > max_words:
                for sub_idx, sub_chunk in enumerate(split_long_text_by_words(sentence_list[i], max_words=max_words)):
                    chunks.append(
                        {
                            "text": sub_chunk,
                            "start_sent_id": i,
                            "end_sent_id": i,
                            "subchunk": sub_idx,
                        }
                    )
                i += 1
                break
            current.append(sentence_list[i])
            total_words += sentence_words
            i += 1
        if current:
            chunks.append(
                {
                    "text": normalize_whitespace(" ".join(current)),
                    "start_sent_id": start_i,
                    "end_sent_id": i - 1,
                    "subchunk": None,
                }
            )
        if overlap_sentences > 0 and i < len(sentence_list):
            i = max(start_i + 1, i - overlap_sentences)
    return chunks


def canonical_label(dataset: str, raw_label: object) -> str | None:
    if raw_label is None:
        return None
    label = normalize_whitespace(raw_label)
    if dataset == "vifactcheck":
        return {"0": "SUPPORTS", "1": "REFUTES", "2": "NEI"}.get(label)
    normalized = normalize_for_hash(label)
    if dataset == "averitec":
        mapping = {
            "supported": "SUPPORTS",
            "refuted": "REFUTES",
            "not enough evidence": "NEI",
            "conflicting evidence cherrypicking": "CONFLICTING",
            "conflicting evidence cherry picking": "CONFLICTING",
            "conflicting evidence cherry-picking": "CONFLICTING",
        }
        return mapping.get(normalized)
    if dataset == "healthver":
        mapping = {
            "supports": "SUPPORTS",
            "support": "SUPPORTS",
            "refutes": "REFUTES",
            "refute": "REFUTES",
            "neutral": "NEI",
            "nei": "NEI",
        }
        return mapping.get(normalized)
    return label.upper() if label else None


def json_safe(value: object) -> object:
    if value is None:
        return None
    if isinstance(value, (str, int, float, bool)):
        return value
    return str(value)