Upload QuestionAndAnswerExtraction.ipynb

Add jupyter script used to create first version of dataset.

QuestionAndAnswerExtraction.ipynb

@@ -0,0 +1,317 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "f31245e6",
+   "metadata": {},
+   "source": [
+    "### Download entries from DIP-Bundestag and put them in a csv to further process them.\n",
+    "\n",
+    "See https://search.dip.bundestag.de/api/v1/swagger-ui/ for the API reference. We only request documents of the type antwort and based on the start and end date provided."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "8af55e90",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import requests\n",
+    "from pprint import pprint\n",
+    "import pandas as pd\n",
+    "from pathlib import Path\n",
+    "from pprint import pprint\n",
+    "from tqdm import tqdm\n",
+    "from concurrent.futures import ThreadPoolExecutor\n",
+    "\n",
+    "DIP_URL = \"https://search.dip.bundestag.de/api/v1/drucksache\"\n",
+    "DIP_TOKEN = \"rgsaY4U.oZRQKUHdJhF9qguHMkwCGIoLaqEcaHjYLF\"\n",
+    "\n",
+    "START_DATE = \"2015-05-07\"\n",
+    "END_DATE = \"2023-07-09\"\n",
+    "\n",
+    "REQUEST_URL = f\"{DIP_URL}?f.drucksachetyp=Antwort&f.datum.start={START_DATE}&f.datum.end={END_DATE}&format=json&apikey={DIP_TOKEN}\"\n",
+    "\n",
+    "df = pd.DataFrame()\n",
+    "docs = []\n",
+    "res = requests.get(REQUEST_URL)\n",
+    "r_json = res.json()\n",
+    "old_cursor = r_json[\"cursor\"]\n",
+    "\n",
+    "with ThreadPoolExecutor(max_workers=10) as pool:\n",
+    "    count = 0\n",
+    "    while True:\n",
+    "        for doc in tqdm(r_json[\"documents\"]):\n",
+    "            docs.append(doc)\n",
+    "            doc_id = doc[\"id\"]\n",
+    "            doc_number = doc[\"fundstelle\"][\"dokumentnummer\"]\n",
+    "            url = doc[\"fundstelle\"][\"pdf_url\"]            \n",
+    "            count += 1\n",
+    "        res = requests.get(f\"{REQUEST_URL}&cursor={old_cursor}\")\n",
+    "        r_json = res.json()\n",
+    "        new_cursor = r_json[\"cursor\"]\n",
+    "        if new_cursor == old_cursor:\n",
+    "            print(\"Found same cursor. No new results.\")\n",
+    "            break\n",
+    "        old_cursor = new_cursor\n",
+    "\n",
+    "\n",
+    "df = df.from_records(docs)\n",
+    "print(f\"Extracted {len(df)} entries.\")\n",
+    "df.to_csv(\"raw_entries.csv\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "58b0055f",
+   "metadata": {},
+   "source": [
+    "### Read back csv written in previous step, and download the associated PDF with each entry"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "1bf8044e",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "\n",
+    "df = pd.read_csv(\"raw_entries.csv\")\n",
+    "\n",
+    "def download_file(download_path:Path,doc_id:str,url:str) -> None:\n",
+    "    r = requests.get(url, allow_redirects=True)\n",
+    "    if r.status_code != 200:\n",
+    "        print(f\"Got status {r.status_code} for url {doc_id} and {url}\")\n",
+    "        return False\n",
+    "    with open(download_path / f\"{doc_id}.pdf\", 'wb') as f:\n",
+    "        f.write(r.content)\n",
+    "    \n",
+    "    return True\n",
+    "\n",
+    "download_path = Path(\"./downloads2/\")\n",
+    "download_path.mkdir(exist_ok=True)\n",
+    "\n",
+    "\n",
+    "for i,row in df.iterrows():\n",
+    "    pdf_url = eval(row[\"fundstelle\"])[\"pdf_url\"]\n",
+    "    success = download_file(download_path,row[\"id\"],pdf_url)\n",
+    "    df.at[i,\"download_success\"]=success\n",
+    "\n",
+    "df.to_csv(\"entries_with_download_status.csv\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "15959ea4",
+   "metadata": {},
+   "source": [
+    "\n",
+    "### Extract the text out of the downloaded pdfs"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "7417bc4e",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import pandas as pd\n",
+    "import sys\n",
+    "import pdftotext\n",
+    "import fitz\n",
+    "import re\n",
+    "from pathlib import Path\n",
+    "from dehyphen import FlairScorer\n",
+    "from dehyphen import format\n",
+    "from tqdm import tqdm\n",
+    "\n",
+    "HEADER_HEIGHT = 78\n",
+    "FOOTER_HEIGHT = 70\n",
+    "\n",
+    "QUESTION_FONT_SIZE = 9.609999656677246\n",
+    "ANSWER_FONT_SIZE = 10.678000450134277\n",
+    "BULLET_POINT_ANSWER_SIZE = 6.0\n",
+    "\n",
+    "OUTPUT_PATH = 'raw_text_blocks.csv'\n",
+    "\n",
+    "scorer = FlairScorer(lang=\"de\")\n",
+    "pattern = r'^\\s*\\d+\\.\\s*' # Matches a number followed by a dot and a space at the beginning of the string\n",
+    "\n",
+    "\n",
+    "def process_text_block(block:dict = {},pdf_path:Path = None,remove_q_numbers: bool = False):\n",
+    "    txt = []\n",
+    "    font = None\n",
+    "\n",
+    "    for line in block.get(\"lines\", []):\n",
+    "        for span in line[\"spans\"]:\n",
+    "            span_txt = span[\"text\"]\n",
+    "            span_font = span[\"font\"]\n",
+    "            span_font_size = span[\"size\"]\n",
+    "            if span_txt==\"\" or span_txt.isspace():\n",
+    "                #print(f\"Found empty string or only spaces in document {pdf_path}\")\n",
+    "                continue\n",
+    "            if font is None:\n",
+    "                span_type = \"Unknown\"\n",
+    "                if span_font_size == QUESTION_FONT_SIZE:\n",
+    "                    span_type = \"Question\"\n",
+    "                    if remove_q_numbers:\n",
+    "                        span_txt = re.sub(pattern,\"\",span_txt)\n",
+    "                elif span_font_size in [BULLET_POINT_ANSWER_SIZE,ANSWER_FONT_SIZE]:\n",
+    "                    span_type = \"Answer\"\n",
+    "                font = (span_font, span_font_size, span_type)\n",
+    "            txt.append(span_txt)\n",
+    "\n",
+    "    if len(txt) > 1:\n",
+    "        txt_joined = \"\\n\".join(txt)\n",
+    "        txt_formatted = format.text_to_format(txt_joined)\n",
+    "        txt_dehyphenated = scorer.dehyphen(txt_formatted)\n",
+    "        txt = format.format_to_text(txt_dehyphenated)\n",
+    "    else:\n",
+    "        if len(txt) == 0:\n",
+    "            txt = \"\"\n",
+    "        else:\n",
+    "            txt = txt[0]\n",
+    "    txt = txt.strip()\n",
+    "    if font is not None:\n",
+    "        result = {\"file\": pdf_path.name, \"txt\": txt, \"font\": font[0], \"size\": font[1], \"type\": font[2]}\n",
+    "    else:\n",
+    "        result = {\"file\": pdf_path.name, \"txt\": \"Error\", \"font\": \"Error\", \"size\":\"Error\", \"type\": \"Error\"}\n",
+    "    return result\n",
+    "\n",
+    "\n",
+    "processed = []\n",
+    "if Path(OUTPUT_PATH).exists():\n",
+    "    df = pd.read_csv(OUTPUT_PATH,sep=\"|\")\n",
+    "    processed = df[\"file\"].values\n",
+    "else:\n",
+    "    df = pd.DataFrame()\n",
+    "\n",
+    "res = []\n",
+    "for pdf_path in tqdm(Path(\"./downloads2\").glob(\"*.pdf\"),desc=\"docs\"):\n",
+    "\n",
+    "    if pdf_path.name in processed:\n",
+    "        print(f\"Found pdf in df: {pdf_path}\")\n",
+    "        continue\n",
+    "\n",
+    "    doc = fitz.open(pdf_path) # open a document\n",
+    "\n",
+    "    for i,page in enumerate(doc):  # iterate the document pages\n",
+    "        #page.draw_rect([0,HEADER_HEIGHT,page.rect.width,page.rect.height - FOOTER_HEIGHT])\n",
+    "        res_raw = page.get_text(\"dict\",clip = [0,HEADER_HEIGHT,page.rect.width,page.rect.height - FOOTER_HEIGHT])\n",
+    "        blocks = res_raw[\"blocks\"] # blocks on page\n",
+    "\n",
+    "        for block in blocks:\n",
+    "            try:\n",
+    "                block_res = process_text_block(block,pdf_path)\n",
+    "                if block_res[\"type\"] != \"Unknown\" and  block_res[\"type\"] != \"Error\":\n",
+    "                    res.append(block_res)\n",
+    "            except Exception as e:\n",
+    "                print(str(e),block)\n",
+    "\n",
+    "\n",
+    "    #print(len(res))\n",
+    "    if len(res) > 100:\n",
+    "        df = pd.DataFrame.from_dict(res)\n",
+    "\n",
+    "        df.to_csv(OUTPUT_PATH, mode='a', header=not Path(OUTPUT_PATH).exists(),index=False,sep=\"|\")\n",
+    "        df = pd.DataFrame()\n",
+    "        res = []\n"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "c9d709c4",
+   "metadata": {},
+   "source": [
+    "### Transform raw text into question / answer tuples"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "98cb494b",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "df_f = pd.read_csv(\"./raw_text_blocks.csv\",sep=\"|\")\n",
+    "print(len(df_f))\n",
+    "files = df_f.groupby('file')\n",
+    "\n",
+    "pairs = []\n",
+    "for i,group in files:\n",
+    "\n",
+    "    i = iter(group.groupby([(group.type != group.type.shift()).cumsum()]))\n",
+    "\n",
+    "    try:\n",
+    "        while True:\n",
+    "            elem1 = next(i)\n",
+    "            if set(elem1[1].type.values) != {\"Question\"}:\n",
+    "                print(\"Broken\")\n",
+    "                continue\n",
+    "            elem2 = next(i)\n",
+    "            if set(elem2[1].type.values) != {\"Answer\"}:\n",
+    "                print(\"Broken\")\n",
+    "                continue\n",
+    "\n",
+    "            pair = {}\n",
+    "            pair[\"question\"] = \"\\n\".join(list(elem1[1].txt.values))\n",
+    "            pair[\"answer\"] = \"\\n\".join(list(elem2[1].txt.values))\n",
+    "            pair[\"doc_id\"] = group.file.unique()[0].split(\".\")[0]\n",
+    "            pairs.append(pair)\n",
+    "    except StopIteration:\n",
+    "        pass\n",
+    "    \n",
+    "df_res = pd.DataFrame.from_records(pairs)\n",
+    "df_res.to_csv(\"final.csv\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "9a816523",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "4a21ca40",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "1ca5572b",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.6"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}