chore: restructured project, updated readme

This commit is contained in:
2025-10-29 09:31:40 +01:00
parent a4460ec17b
commit ee62c65ae7
70 changed files with 8518 additions and 100 deletions

373
src/parsers/word_parser.py Normal file
View File

@@ -0,0 +1,373 @@
import zipfile
from typing import Any, Optional
import fitz # PyMuPDF
import pandas as pd
from bs4 import BeautifulSoup
from docx import Document
from src.core.models import Book, SemapDocument
from src.shared.logging import log
def word_docx_to_csv(path: str) -> list[pd.DataFrame]:
doc = Document(path)
tables = doc.tables
m_data = []
for table in tables:
data = []
for row in table.rows:
row_data: list[Any] = []
for cell in row.cells:
text = cell.text
text = text.replace("\n", "")
row_data.append(text)
# if text == "Ihr Fach:":
# row_data.append(get_fach(path))
data.append(row_data)
df = pd.DataFrame(data)
df.columns = df.iloc[0]
df = df.iloc[1:]
m_data.append(df)
return m_data
def get_fach(path: str) -> Optional[str]:
document = zipfile.ZipFile(path)
xml_data = document.read("word/document.xml")
document.close()
soup = BeautifulSoup(xml_data, "xml")
# text we need is in <w:p w14:paraId="12456A32" ... > -> w:r -> w:t
paragraphs = soup.find_all("w:p")
for para in paragraphs:
para_id = para.get("w14:paraId")
if para_id == "12456A32":
# get the data in the w:t
for run in para.find_all("w:r"):
data = run.find("w:t")
if data and data.contents:
return data.contents[0]
return None
def makeDict() -> dict[str, Optional[str]]:
return {
"work_author": None,
"section_author": None,
"year": None,
"edition": None,
"work_title": None,
"chapter_title": None,
"location": None,
"publisher": None,
"signature": None,
"issue": None,
"pages": None,
"isbn": None,
"type": None,
}
def tuple_to_dict(tlist: tuple, type: str) -> list[dict[str, Optional[str]]]:
ret: list[dict[str, Optional[str]]] = []
for line in tlist:
data = makeDict()
if type == "Monografien":
data["type"] = type
data["work_author"] = line[0]
data["year"] = line[1]
data["edition"] = line[2]
data["work_title"] = line[3]
data["location"] = line[4]
data["publisher"] = line[5]
data["signature"] = line[6]
data["pages"] = line[7]
elif type == "Herausgeberwerke":
data["type"] = type
data["section_author"] = line[0]
data["year"] = line[1]
data["edition"] = line[2]
data["chapter_title"] = line[3]
data["work_author"] = line[4]
data["work_title"] = line[5]
data["location"] = line[6]
data["publisher"] = line[7]
data["signature"] = line[9]
data["pages"] = line[8]
elif type == "Zeitschriftenaufsätze":
data["type"] = type
data["section_author"] = line[0]
data["year"] = line[1]
data["issue"] = line[2]
data["chapter_title"] = line[3]
data["work_title"] = line[4]
data["location"] = line[5]
data["publisher"] = line[6]
data["signature"] = line[8]
data["pages"] = line[7]
ret.append(data)
return ret
def elsa_word_to_csv(path: str) -> tuple[list[dict[str, Optional[str]]], str]:
doc = Document(path)
# # print all lines in doc
doctype = [para.text for para in doc.paragraphs if para.text != ""][-1]
tuples = {
"Monografien": ("", "", "", "", "", "", "", "", ""),
"Herausgeberwerke": ("", "", "", "", "", "", "", "", "", "", ""),
"Zeitschriftenaufsätze": ("", "", "", "", "", "", "", "", "", ""),
}
tables = doc.tables
m_data: list[pd.DataFrame] = []
for table in tables:
data: list[list[str]] = []
for row in table.rows:
row_data: list[str] = []
for cell in row.cells:
text = cell.text
text = text.replace("\n", "")
text = text.replace("\u2002", "")
row_data.append(text)
data.append(row_data)
df = pd.DataFrame(data)
df.columns = df.iloc[0]
df = df.iloc[1:]
m_data.append(df)
df = m_data[0]
# split df to rows
data = [
row for row in df.itertuples(index=False, name=None) if row != tuples[doctype]
]
# log.debug(data)
return tuple_to_dict(data, doctype), doctype
def word_to_semap(word_path: str, ai: bool = True) -> SemapDocument:
log.info("Parsing Word Document {}", word_path)
semap = SemapDocument()
df = word_docx_to_csv(word_path)
apparatdata = df[0]
apparatdata = apparatdata.to_dict()
keys = list(apparatdata.keys())
# print(apparatdata, keys)
appdata = {keys[i]: keys[i + 1] for i in range(0, len(keys) - 1, 2)}
semap.phoneNumber = appdata["Telefon:"]
semap.subject = appdata["Ihr Fach:"]
semap.mail = appdata["Mailadresse:"]
semap.personName = ",".join(appdata["Ihr Name und Titel:"].split(",")[:-1])
semap.personTitle = ",".join(appdata["Ihr Name und Titel:"].split(",")[-1:]).strip()
apparatdata = df[1]
apparatdata = apparatdata.to_dict()
keys = list(apparatdata.keys())
appdata = {keys[i]: keys[i + 1] for i in range(0, len(keys), 2)}
semap.title = appdata["Veranstaltung:"]
semap.semester = appdata["Semester:"]
if ai:
semap.renameSemester
semap.nameSetter
books = df[2]
booklist = []
for i in range(len(books)):
if books.iloc[i].isnull().all():
continue
data = books.iloc[i].to_dict()
book = Book()
book.from_dict(data)
if book.is_empty:
continue
elif not book.has_signature:
continue
else:
booklist.append(book)
log.info("Found {} books", len(booklist))
semap.books = booklist
return semap
def pdf_to_semap(pdf_path: str, ai: bool = True) -> SemapDocument:
"""
Parse a Semesterapparat PDF like the sample you provided and return a SemapDocument.
- No external programs, only PyMuPDF.
- Robust to multi-line field values (e.g., hyphenated emails) and multi-line table cells.
- Works across multiple pages; headers only need to exist on the first page.
"""
doc = fitz.open(pdf_path)
semap = SemapDocument()
# ---------- helpers ----------
def _join_tokens(tokens: list[str]) -> str:
"""Join tokens, preserving hyphen/URL joins across line wraps."""
parts = []
for tok in tokens:
if parts and (
parts[-1].endswith("-")
or parts[-1].endswith("/")
or parts[-1].endswith(":")
):
parts[-1] = parts[-1] + tok # no space after '-', '/' or ':'
else:
parts.append(tok)
return " ".join(parts).strip()
def _extract_row_values_multiline(
page, labels: list[str], y_window: float = 24
) -> dict[str, str]:
"""For a row of inline labels (e.g., Name/Fach/Telefon/Mail), grab text to the right of each label."""
rects = []
for lab in labels:
hits = page.search_for(lab)
if hits:
rects.append((lab, hits[0]))
if not rects:
return {}
rects.sort(key=lambda t: t[1].x0)
words = page.get_text("words")
out = {}
for i, (lab, r) in enumerate(rects):
x0 = r.x1 + 1
x1 = rects[i + 1][1].x0 - 1 if i + 1 < len(rects) else page.rect.width - 5
y0 = r.y0 - 3
y1 = r.y0 + y_window
toks = [w for w in words if x0 <= w[0] <= x1 and y0 <= w[1] <= y1]
toks.sort(key=lambda w: (w[1], w[0])) # line, then x
out[lab] = _join_tokens([w[4] for w in toks])
return out
def _compute_columns_from_headers(page0):
"""Find column headers (once) and derive column centers + header baseline."""
headers = [
("Autorenname(n):", "Autorenname(n):Nachname, Vorname"),
("Jahr/Auflage", "Jahr/Auflage"),
("Titel", "Titel"),
("Ort und Verlag", "Ort und Verlag"),
("Standnummer", "Standnummer"),
("Interne Vermerke", "Interne Vermerke"),
]
found = []
for label, canon in headers:
rects = [
r for r in page0.search_for(label) if r.y0 > 200
] # skip top-of-form duplicates
if rects:
found.append((canon, rects[0]))
found.sort(key=lambda t: t[1].x0)
cols = [(canon, r.x0, r.x1, (r.x0 + r.x1) / 2.0) for canon, r in found]
header_y = min(r.y0 for _, r in found) if found else 0
return cols, header_y
def _extract_table_rows_from_page(
page, cols, header_y, y_top_margin=5, y_bottom_margin=40, y_tol=26.0
):
"""
Group words into logical rows (tolerant to wrapped lines), then map each word
to the nearest column by x-center and join tokens per column.
"""
words = [
w
for w in page.get_text("words")
if w[1] > header_y + y_top_margin
and w[3] < page.rect.height - y_bottom_margin
]
# group into row bands by y (tolerance big enough to capture wrapped lines, but below next row gap)
rows = []
for w in sorted(words, key=lambda w: w[1]):
y = w[1]
for row in rows:
if abs(row["y_mean"] - y) <= y_tol:
row["ys"].append(y)
row["y_mean"] = sum(row["ys"]) / len(row["ys"])
row["words"].append(w)
break
else:
rows.append({"y_mean": y, "ys": [y], "words": [w]})
# map to columns + join
joined_rows = []
for row in rows:
rowdict = {canon: "" for canon, *_ in cols}
words_by_col = {canon: [] for canon, *_ in cols}
for w in sorted(row["words"], key=lambda w: (w[1], w[0])):
xmid = (w[0] + w[2]) / 2.0
canon = min(cols, key=lambda c: abs(xmid - c[3]))[0]
words_by_col[canon].append(w[4])
for canon, toks in words_by_col.items():
rowdict[canon] = _join_tokens(toks)
if any(v for v in rowdict.values()):
joined_rows.append(rowdict)
return joined_rows
# ---------- top-of-form fields ----------
p0 = doc[0]
row1 = _extract_row_values_multiline(
p0,
["Ihr Name und Titel:", "Ihr Fach:", "Telefon:", "Mailadresse:"],
y_window=22,
)
row2 = _extract_row_values_multiline(
p0, ["Veranstaltung:", "Semester:"], y_window=20
)
name_title = row1.get("Ihr Name und Titel:", "") or ""
semap.subject = row1.get("Ihr Fach:", None)
semap.phoneNumber = row1.get("Telefon:", None) # keep as-is (string like "682-308")
semap.mail = row1.get("Mailadresse:", None)
semap.personName = ",".join(name_title.split(",")[:-1]) if name_title else None
semap.personTitle = (
",".join(name_title.split(",")[-1:]).strip() if name_title else None
)
semap.title = row2.get("Veranstaltung:", None)
semap.semester = row2.get("Semester:", None)
# ---------- table extraction (all pages) ----------
cols, header_y = _compute_columns_from_headers(p0)
all_rows: list[dict[str, Any]] = []
for pn in range(len(doc)):
all_rows.extend(_extract_table_rows_from_page(doc[pn], cols, header_y))
# drop the sub-header line "Nachname, Vorname" etc.
filtered = []
for r in all_rows:
if r.get("Autorenname(n):Nachname, Vorname", "").strip() in (
"",
"Nachname, Vorname",
):
# skip if it's just the sub-header line
if all(not r[c] for c in r if c != "Autorenname(n):Nachname, Vorname"):
continue
filtered.append(r)
# build Book objects (same filters as your word parser)
booklist: list[Book] = []
for row in filtered:
b = Book()
b.from_dict(row)
if b.is_empty:
continue
if not b.has_signature:
continue
booklist.append(b)
semap.books = booklist
# keep parity with your post-processing
if ai:
_ = semap.renameSemester
_ = semap.nameSetter
return semap
if __name__ == "__main__":
else_df = pdf_to_semap("C:/Users/aky547/Dokumente/testsemap.pdf")
# print(else_df)