{"id":812,"date":"2026-04-29T15:08:42","date_gmt":"2026-04-29T07:08:42","guid":{"rendered":"https:\/\/connectword.dpdns.org\/?p=812"},"modified":"2026-04-29T15:08:42","modified_gmt":"2026-04-29T07:08:42","slug":"a-coding-implementation-on-document-parsing-benchmarking-with-llamaindex-parsebench-using-python-hugging-face-and-evaluation-metrics","status":"publish","type":"post","link":"https:\/\/connectword.dpdns.org\/?p=812","title":{"rendered":"A Coding Implementation on Document Parsing Benchmarking with LlamaIndex ParseBench Using Python, Hugging Face, and Evaluation Metrics"},"content":{"rendered":"

In this tutorial, we explore how to use the ParseBench<\/strong><\/a> dataset to evaluate document parsing systems in a structured, practical way. We begin by loading the dataset directly from Hugging Face, inspecting its multiple dimensions, such as text, tables, charts, and layout, and transforming it into a unified dataframe for deeper analysis. As we progress, we identify key fields, detect linked PDFs, and build a lightweight baseline using PyMuPDF to extract and compare text. Throughout the process, we focus on creating a flexible pipeline that allows us to understand the dataset schema, evaluate parsing quality, and prepare inputs for more advanced OCR or vision-language models.<\/p>\n

\n
\n
\n
\n
<\/div>\n
<\/div>\n
<\/div>\n<\/div>\n
Copy Code<\/span>Copied<\/span>Use a different Browser<\/span><\/a><\/div>\n<\/div>\n
!pip install -q -U datasets huggingface_hub pandas matplotlib rich pymupdf rapidfuzz tqdm\n\n\nimport json, re, textwrap, random, math\nfrom pathlib import Path\nfrom collections import Counter\nimport pandas as pd\nimport matplotlib.pyplot as plt\nfrom tqdm.auto import tqdm\nfrom rich.console import Console\nfrom rich.table import Table\nfrom rich.panel import Panel\nfrom huggingface_hub import hf_hub_download, list_repo_files\nfrom rapidfuzz import fuzz\nimport fitz\n\n\nconsole = Console()\nDATASET_ID = \"llamaindex\/ParseBench\"\nWORKDIR = Path(\"\/content\/parsebench_tutorial\")\nWORKDIR.mkdir(parents=True, exist_ok=True)\n\n\nconsole.print(Panel.fit(\"Advanced ParseBench Tutorial on Google Colab\", style=\"bold green\"))\n\n\nfiles = list_repo_files(DATASET_ID, repo_type=\"dataset\")\njsonl_files = [f for f in files if f.endswith(\".jsonl\")]\npdf_files = [f for f in files if f.endswith(\".pdf\")]\n\n\nconsole.print(f\"Found {len(jsonl_files)} JSONL files\")\nconsole.print(f\"Found {len(pdf_files)} PDF files\")\n\n\ntable = Table(title=\"ParseBench JSONL Files\")\ntable.add_column(\"File\")\ntable.add_column(\"Dimension\")\nfor f in jsonl_files:\n   table.add_row(f, Path(f).stem)\nconsole.print(table)<\/code><\/pre>\n<\/div>\n<\/div>\n
We install all required libraries and set up our working environment for the tutorial. We initialize the dataset source and prepare a workspace to store all outputs. We also fetch and list all JSONL and PDF files from the ParseBench repository to understand the dataset structure.<\/p>\n
\n
\n
\n
\n
<\/div>\n
<\/div>\n
<\/div>\n<\/div>\n
Copy Code<\/span>Copied<\/span>Use a different Browser<\/span><\/a><\/div>\n<\/div>\n
def load_jsonl_from_hf(filename, max_rows=None):\n   path = hf_hub_download(repo_id=DATASET_ID, filename=filename, repo_type=\"dataset\")\n   rows = []\n   with open(path, \"r\", encoding=\"utf-8\") as fp:\n       for i, line in enumerate(fp):\n           if max_rows and i >= max_rows:\n               break\n           line = line.strip()\n           if line:\n               rows.append(json.loads(line))\n   return rows, path\n\n\ndef flatten_dict(d, parent_key=\"\", sep=\".\"):\n   items = {}\n   if isinstance(d, dict):\n       for k, v in d.items():\n           new_key = f\"{parent_key}{sep}{k}\" if parent_key else str(k)\n           if isinstance(v, dict):\n               items.update(flatten_dict(v, new_key, sep=sep))\n           else:\n               items[new_key] = v\n   return items\n\n\ndimension_data = {}\nfor jf in jsonl_files:\n   rows, local_path = load_jsonl_from_hf(jf)\n   dimension_data[Path(jf).stem] = rows\n   console.print(f\"{jf}: {len(rows)} examples loaded\")\n\n\nsummary_rows = []\nfor dim, rows in dimension_data.items():\n   keys = Counter()\n   for r in rows[:100]:\n       keys.update(flatten_dict(r).keys())\n   summary_rows.append({\n       \"dimension\": dim,\n       \"examples\": len(rows),\n       \"top_fields\": \", \".join([k for k, _ in keys.most_common(12)])\n   })\n\n\nsummary_df = pd.DataFrame(summary_rows)\ndisplay(summary_df)\n\n\nplt.figure(figsize=(10, 5))\nplt.bar(summary_df[\"dimension\"], summary_df[\"examples\"])\nplt.title(\"ParseBench Examples by Dimension\")\nplt.xlabel(\"Dimension\")\nplt.ylabel(\"Number of Examples\")\nplt.xticks(rotation=30, ha=\"right\")\nplt.show()\n\n\nfor dim, rows in dimension_data.items():\n   console.print(Panel.fit(f\"Sample schema for {dim}\", style=\"bold cyan\"))\n   if rows:\n       console.print(json.dumps(rows[0], indent=2)[:3000])<\/code><\/pre>\n<\/div>\n<\/div>\n
We load the JSONL files from the dataset and convert them into usable Python objects. We flatten nested structures to analyze them easily in a tabular format. We also summarize each dimension and visualize the distribution of examples across different parsing tasks.<\/p>\n
\n
\n
\n
\n
<\/div>\n
<\/div>\n
<\/div>\n<\/div>\n
Copy Code<\/span>Copied<\/span>Use a different Browser<\/span><\/a><\/div>\n<\/div>\n
all_records = []\nfor dim, rows in dimension_data.items():\n   for i, r in enumerate(rows):\n       flat = flatten_dict(r)\n       flat[\"_dimension\"] = dim\n       flat[\"_row_id\"] = i\n       all_records.append(flat)\n\n\ndf = pd.DataFrame(all_records)\nconsole.print(f\"Combined dataframe shape: {df.shape}\")\ndisplay(df.head())\n\n\nmissing_report = []\nfor col in df.columns:\n   missing_report.append({\n       \"column\": col,\n       \"non_null\": int(df[col].notna().sum()),\n       \"missing\": int(df[col].isna().sum()),\n       \"coverage_pct\": round(100 * df[col].notna().mean(), 2)\n   })\n\n\nmissing_df = pd.DataFrame(missing_report).sort_values(\"coverage_pct\", ascending=False)\ndisplay(missing_df.head(40))\n\n\ndef find_candidate_columns(df, keywords):\n   cols = []\n   for c in df.columns:\n       lc = c.lower()\n       if any(k.lower() in lc for k in keywords):\n           cols.append(c)\n   return cols\n\n\ndoc_cols = find_candidate_columns(df, [\"doc\", \"pdf\", \"file\", \"path\", \"source\", \"image\"])\ntext_cols = find_candidate_columns(df, [\"text\", \"content\", \"markdown\", \"ground\", \"answer\", \"expected\", \"target\", \"reference\"])\nrule_cols = find_candidate_columns(df, [\"rule\", \"check\", \"assert\", \"criteria\", \"question\", \"prompt\"])\nbbox_cols = find_candidate_columns(df, [\"bbox\", \"box\", \"polygon\", \"coordinates\", \"layout\"])\n\n\nconsole.print(\"[bold]Possible document columns:[\/bold]\", doc_cols[:30])\nconsole.print(\"[bold]Possible text\/reference columns:[\/bold]\", text_cols[:30])\nconsole.print(\"[bold]Possible rule\/question columns:[\/bold]\", rule_cols[:30])\nconsole.print(\"[bold]Possible layout columns:[\/bold]\", bbox_cols[:30])<\/code><\/pre>\n<\/div>\n<\/div>\n
We combine all parsed records into a single dataframe for unified analysis. We evaluate missing values and identify which fields are most informative across the dataset. We also detect candidate columns related to documents, text, rules, and layout to guide downstream processing.<\/p>\n
\n
\n
\n
\n
<\/div>\n
<\/div>\n
<\/div>\n<\/div>\n
Copy Code<\/span>Copied<\/span>Use a different Browser<\/span><\/a><\/div>\n<\/div>\n
def pick_first_existing(row, candidates):\n   for c in candidates:\n       if c in row and pd.notna(row[c]):\n           value = row[c]\n           if isinstance(value, str) and value.strip():\n               return value\n           if not isinstance(value, str):\n               return value\n   return None\n\n\ndef normalize_text(x):\n   if x is None or (isinstance(x, float) and math.isnan(x)):\n       return \"\"\n   x = str(x)\n   x = re.sub(r\"s+\", \" \", x)\n   return x.strip().lower()\n\n\ndef simple_text_similarity(a, b):\n   a = normalize_text(a)\n   b = normalize_text(b)\n   if not a or not b:\n       return None\n   return fuzz.token_set_ratio(a, b) \/ 100\n\n\ndef locate_pdf_path(value):\n   if value is None:\n       return None\n   value = str(value)\n   candidates = []\n   if value.endswith(\".pdf\"):\n       candidates.append(value)\n       candidates.extend([f for f in pdf_files if f.endswith(value.split(\"\/\")[-1])])\n   else:\n       candidates.extend([\n           f for f in pdf_files\n           if value in f or Path(f).stem in value or value in Path(f).stem\n       ])\n   return candidates[0] if candidates else None\n\n\ndef extract_pdf_text_from_hf(pdf_repo_path, max_pages=2):\n   local_pdf = hf_hub_download(repo_id=DATASET_ID, filename=pdf_repo_path, repo_type=\"dataset\")\n   doc = fitz.open(local_pdf)\n   texts = []\n   for page_idx in range(min(max_pages, len(doc))):\n       texts.append(doc[page_idx].get_text(\"text\"))\n   doc.close()\n   return \"n\".join(texts), local_pdf\n\n\ndef render_pdf_first_page(pdf_repo_path, zoom=2):\n   local_pdf = hf_hub_download(repo_id=DATASET_ID, filename=pdf_repo_path, repo_type=\"dataset\")\n   doc = fitz.open(local_pdf)\n   page = doc[0]\n   pix = page.get_pixmap(matrix=fitz.Matrix(zoom, zoom))\n   out_path = WORKDIR \/ (Path(pdf_repo_path).stem + \"_page1.png\")\n   pix.save(out_path)\n   doc.close()\n   return out_path\n\n\nsample_records = df.sample(min(25, len(df)), random_state=42).to_dict(\"records\")\npdf_candidates = []\n\n\nfor row in sample_records:\n   for c in doc_cols:\n       pdf_path = locate_pdf_path(row.get(c))\n       if pdf_path:\n           pdf_candidates.append((row[\"_dimension\"], row[\"_row_id\"], pdf_path))\n           break\n\n\npdf_candidates = list(dict.fromkeys(pdf_candidates))\nconsole.print(f\"Detected {len(pdf_candidates)} PDF-linked sampled records\")\n\n\nif pdf_candidates:\n   dim, row_id, pdf_path = pdf_candidates[0]\n   console.print(Panel.fit(f\"Rendering sample PDFnDimension: {dim}nRow: {row_id}nPDF: {pdf_path}\", style=\"bold yellow\"))\n   image_path = render_pdf_first_page(pdf_path)\n   img = plt.imread(image_path)\n   plt.figure(figsize=(10, 12))\n   plt.imshow(img)\n   plt.axis(\"off\")\n   plt.title(f\"{dim}: {Path(pdf_path).name}\")\n   plt.show()\nelse:\n   console.print(\"[yellow]No PDF-linked rows were detected from the sample.[\/yellow]\")<\/code><\/pre>\n<\/div>\n<\/div>\n
We define helper functions for text normalization, similarity scoring, and PDF handling. We locate and download PDF files associated with dataset entries and extract their textual content. We also provide a sample PDF page for visual inspection of the document structure.<\/p>\n
\n
\n
\n
\n
<\/div>\n
<\/div>\n
<\/div>\n<\/div>\n
Copy Code<\/span>Copied<\/span>Use a different Browser<\/span><\/a><\/div>\n<\/div>\n
preferred_gt_cols = [\n   c for c in text_cols\n   if any(k in c.lower() for k in [\"ground\", \"expected\", \"target\", \"answer\", \"content\", \"text\", \"markdown\", \"reference\"])\n]\n\n\nevaluation_rows = []\neval_sample = df.sample(min(50, len(df)), random_state=7).to_dict(\"records\")\n\n\nfor row in tqdm(eval_sample, desc=\"Running lightweight PDF text extraction baseline\"):\n   pdf_path = None\n   for c in doc_cols:\n       pdf_path = locate_pdf_path(row.get(c))\n       if pdf_path:\n           break\n\n\n   if not pdf_path:\n       evaluation_rows.append({\n           \"dimension\": row.get(\"_dimension\"),\n           \"row_id\": row.get(\"_row_id\"),\n           \"pdf\": None,\n           \"ground_truth_column\": None,\n           \"similarity_score\": None,\n           \"status\": \"no_pdf_detected\"\n       })\n       continue\n\n\n   gt_col = None\n   gt = None\n   for c in preferred_gt_cols:\n       if c in row and pd.notna(row[c]):\n           gt_col = c\n           gt = row[c]\n           break\n\n\n   if gt is None:\n       evaluation_rows.append({\n           \"dimension\": row.get(\"_dimension\"),\n           \"row_id\": row.get(\"_row_id\"),\n           \"pdf\": pdf_path,\n           \"ground_truth_column\": None,\n           \"similarity_score\": None,\n           \"status\": \"no_reference_detected\"\n       })\n       continue\n\n\n   try:\n       extracted, local_pdf = extract_pdf_text_from_hf(pdf_path, max_pages=2)\n       score = simple_text_similarity(extracted, gt)\n       evaluation_rows.append({\n           \"dimension\": row.get(\"_dimension\"),\n           \"row_id\": row.get(\"_row_id\"),\n           \"pdf\": pdf_path,\n           \"ground_truth_column\": gt_col,\n           \"similarity_score\": score,\n           \"extracted_chars\": len(extracted),\n           \"ground_truth_chars\": len(str(gt)),\n           \"status\": \"scored\"\n       })\n   except Exception as e:\n       evaluation_rows.append({\n           \"dimension\": row.get(\"_dimension\"),\n           \"row_id\": row.get(\"_row_id\"),\n           \"pdf\": pdf_path,\n           \"ground_truth_column\": gt_col,\n           \"similarity_score\": None,\n           \"status\": \"error\",\n           \"error\": str(e)\n       })\n\n\neval_df = pd.DataFrame(evaluation_rows)\n\n\nif eval_df.empty:\n   eval_df = pd.DataFrame(columns=[\n       \"dimension\", \"row_id\", \"pdf\", \"ground_truth_column\",\n       \"similarity_score\", \"extracted_chars\", \"ground_truth_chars\",\n       \"status\", \"error\"\n   ])\n\n\ndisplay(eval_df.head(30))\n\n\nif \"status\" in eval_df.columns:\n   display(eval_df[\"status\"].value_counts().reset_index().rename(columns={\"index\": \"status\", \"status\": \"count\"}))\n\n\nif not eval_df.empty and \"similarity_score\" in eval_df.columns:\n   valid_eval = eval_df.dropna(subset=[\"similarity_score\"])\n\n\n   if len(valid_eval):\n       console.print(f\"Average lightweight text similarity: {valid_eval['similarity_score'].mean():.3f}\")\n\n\n       plt.figure(figsize=(8, 5))\n       plt.hist(valid_eval[\"similarity_score\"], bins=10)\n       plt.title(\"Lightweight Baseline Similarity Distribution\")\n       plt.xlabel(\"RapidFuzz Token Set Similarity\")\n       plt.ylabel(\"Count\")\n       plt.show()\n\n\n       per_dim = valid_eval.groupby(\"dimension\")[\"similarity_score\"].mean().reset_index()\n       display(per_dim)\n\n\n       plt.figure(figsize=(9, 5))\n       plt.bar(per_dim[\"dimension\"], per_dim[\"similarity_score\"])\n       plt.title(\"Average Baseline Similarity by Dimension\")\n       plt.xlabel(\"Dimension\")\n       plt.ylabel(\"Average Similarity\")\n       plt.xticks(rotation=30, ha=\"right\")\n       plt.show()\n   else:\n       console.print(\"[yellow]No valid similarity scores were produced. This usually means sampled rows did not contain both detectable PDFs and reference text.[\/yellow]\")\nelse:\n   console.print(\"[yellow]No similarity_score column found.[\/yellow]\")<\/code><\/pre>\n<\/div>\n<\/div>\n
We run a lightweight evaluation pipeline by comparing extracted text with available reference fields. We compute similarity scores and analyze how well simple extraction performs across different dimensions. We also visualize the results to understand performance trends and limitations.<\/p>\n
\n
\n
\n
\n
<\/div>\n
<\/div>\n
<\/div>\n<\/div>\n
Copy Code<\/span>Copied<\/span>Use a different Browser<\/span><\/a><\/div>\n<\/div>\n
def inspect_dimension(dimension_name, n=3):\n   rows = dimension_data.get(dimension_name, [])\n   console.print(Panel.fit(f\"Inspecting {dimension_name}: {len(rows)} rows\", style=\"bold magenta\"))\n   for idx, row in enumerate(rows[:n]):\n       console.print(f\"n[bold]Example {idx}[\/bold]\")\n       console.print(json.dumps(row, indent=2)[:2500])\n\n\nfor dim in list(dimension_data.keys())[:5]:\n   inspect_dimension(dim, n=1)\n\n\ndef make_parsebench_subset(dimension=None, n=20, seed=123):\n   subset = df.copy()\n   if dimension:\n       subset = subset[subset[\"_dimension\"] == dimension]\n   if len(subset) == 0:\n       return subset\n   return subset.sample(min(n, len(subset)), random_state=seed)\n\n\nsubset = make_parsebench_subset(n=20)\ndisplay(subset.head())\n\n\ndef create_llm_parser_prompt(row):\n   dimension = row.get(\"_dimension\", \"unknown\")\n   candidate_truth = pick_first_existing(row, preferred_gt_cols)\n   rule_hint = pick_first_existing(row, rule_cols)\n\n\n   prompt = f\"\"\"\nYou are evaluating a document parser on ParseBench.\n\n\nDimension:\n{dimension}\n\n\nTask:\nParse the PDF page into a structured representation that preserves the information needed for agentic workflows.\n\n\nRelevant benchmark hint or rule:\n{rule_hint if rule_hint is not None else \"No obvious rule field detected.\"}\n\n\nReference field preview:\n{str(candidate_truth)[:1000] if candidate_truth is not None else \"No obvious reference field detected.\"}\n\n\nReturn:\n1. Markdown representation\n2. Extracted tables as JSON arrays when tables exist\n3. Extracted chart values as JSON when charts exist\n4. Layout-sensitive notes when visual grounding matters\n\"\"\"\n   return textwrap.dedent(prompt).strip()\n\n\nprompt_examples = []\nif len(subset):\n   for _, row in subset.head(3).iterrows():\n       prompt_examples.append(create_llm_parser_prompt(row.to_dict()))\n\n\nif prompt_examples:\n   console.print(Panel.fit(\"Example prompt for testing an external OCR or VLM parser\", style=\"bold blue\"))\n   console.print(prompt_examples[0])\nelse:\n   console.print(\"[yellow]No prompt examples could be created because the subset is empty.[\/yellow]\")\n\n\ndef compare_parser_outputs(reference, candidate):\n   return {\n       \"token_set_similarity\": simple_text_similarity(reference, candidate),\n       \"partial_ratio\": fuzz.partial_ratio(normalize_text(reference), normalize_text(candidate)) \/ 100 if reference and candidate else None,\n       \"candidate_length\": len(str(candidate)) if candidate else 0,\n       \"reference_length\": len(str(reference)) if reference else 0\n   }\n\n\nif not eval_df.empty and \"similarity_score\" in eval_df.columns:\n   scored_eval = eval_df.dropna(subset=[\"similarity_score\"])\n\n\n   if len(scored_eval):\n       best = scored_eval.sort_values(\"similarity_score\", ascending=False).head(1)\n       worst = scored_eval.sort_values(\"similarity_score\", ascending=True).head(1)\n\n\n       console.print(Panel.fit(\"Best lightweight baseline example\", style=\"bold green\"))\n       display(best)\n\n\n       console.print(Panel.fit(\"Worst lightweight baseline example\", style=\"bold red\"))\n       display(worst)\n   else:\n       console.print(\"[yellow]No valid similarity scores were available for best\/worst comparison.[\/yellow]\")\n\n\noutput_path = WORKDIR \/ \"parsebench_flattened_sample.csv\"\ndf.head(500).to_csv(output_path, index=False)\nconsole.print(f\"Saved flattened sample to: {output_path}\")\n\n\nconsole.print(Panel.fit(\"\"\"\nTutorial complete.\n\n\nWhat we build:\n1. Load ParseBench files directly from Hugging Face.\n2. Inspect benchmark dimensions and schemas.\n3. Flatten records into a dataframe.\n4. Detect linked PDFs and render sample pages when possible.\n5. Run a lightweight PyMuPDF extraction baseline.\n6. Score extracted text when reference fields are available.\n7. Generate reusable prompts for OCR, VLM, and document parser evaluation.\n\"\"\", style=\"bold green\"))<\/code><\/pre>\n<\/div>\n<\/div>\n
We inspect dataset samples and create subsets for experimentation. We generate structured prompts for evaluating external parsing systems, such as OCR and vision-language models. Also, we compare outputs, identify best and worst cases, and save processed data for future use.<\/p>\n
In conclusion, we built a complete workflow that allows us to analyze, evaluate, and experiment with document parsing using the ParseBench dataset. We extracted and compared textual content and also generated structured prompts for testing external parsing systems, such as OCR engines and VLMs. This approach helps us move beyond simple text extraction and toward building agent-ready representations that preserve structure, layout, and semantic meaning. Also, we established a strong foundation that we can extend further for benchmarking, improving parsing models, and integrating document understanding into real-world AI pipelines.<\/p>\n
\n
Check out\u00a0the\u00a0Full Codes here<\/a><\/strong>.\u00a0<\/strong>Also,\u00a0feel free to follow us on\u00a0Twitter<\/mark><\/a><\/strong>\u00a0and don\u2019t forget to join our\u00a0130k+ ML SubReddit<\/a><\/strong>\u00a0and Subscribe to\u00a0our Newsletter<\/a><\/strong>. Wait! are you on telegram?\u00a0now you can join us on telegram as well.<\/a><\/strong><\/p>\n
Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.?\u00a0Connect with us<\/mark><\/a><\/strong><\/p>\n
The post A Coding Implementation on Document Parsing Benchmarking with LlamaIndex ParseBench Using Python, Hugging Face, and Evaluation Metrics<\/a> appeared first on MarkTechPost<\/a>.<\/p>","protected":false},"excerpt":{"rendered":"
In this tutorial, we explore h…<\/p>\n","protected":false},"author":1,"featured_media":29,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-812","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=\/wp\/v2\/posts\/812","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=812"}],"version-history":[{"count":0,"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=\/wp\/v2\/posts\/812\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=\/wp\/v2\/media\/29"}],"wp:attachment":[{"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=812"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=812"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/connectword.dpdns.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=812"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}