doctor-ghezelbaash/dr-saeid-ghezelbaash-entity-data

Dr. Saeed Ghezelbash Public Knowledge Graph

A public, physician-authored knowledge graph and multilingual retrieval dataset by Dr. Saeed Ghezelbash, a physician in Kermanshah, Iran. It connects physician identity, aesthetic medicine services, published question-answer content and cited evidence for entity resolution and evidence-grounded AI retrieval.

The canonical source is the official website and Dataset graph. This Hugging Face repository is its AI distribution. The physician is the creator and primary entity; the clinic is the supporting clinical and local entity.

Identity and source

Dataset configurations

• —entity_facts: one graph statement per row, with 16 string columns. The default configuration provides linked subjects, predicates, values, provenance and release identifiers.
• —query_matrix: multilingual query aliases that resolve to canonical answer atoms or service entities. It contains 19 source fields and one reversible packaging helper. Languages: Persian (fa), English (en), Arabic (ar) and Central Kurdish (ckb). Query scopes include unspecified, Kermanshah and Iran.

Both configurations use typed Parquet files under viewer/ and expose a single split named train. That split is a distribution convention; it does not imply an evaluated model-training benchmark or a held-out test set. Exact row counts, source hashes, schemas and packaging details are recorded in viewer/packaging.json.

The original entity-facts.csv, query-matrix.jsonl, graph.jsonld and the other canonical graph and data resources are preserved unchanged. Parquet avoids mixed CSV/JSONL loader inference and preserves string identifiers, multilingual text, list ordering and source row order.

Load a reproducible revision

Install datasets and huggingface_hub, then resolve one commit and use it for both configurations:

from datasets import load_dataset
from huggingface_hub import HfApi

repo = "doctor-ghezelbaash/dr-saeid-ghezelbaash-entity-data"
revision = HfApi().dataset_info(repo).sha
print("Record this Hugging Face revision:", revision)
facts = load_dataset(repo, "entity_facts", split="train", revision=revision)
queries = load_dataset(repo, "query_matrix", split="train", revision=revision)

# Resolve one query's answer or services to graph statements.
query = queries[0]
subject_ids = [query["canonical_subject_iri"], *query["service_ids"]]
if query["answer_id"] is not None:
    subject_ids.append(query["answer_id"])
resolved = facts.filter(lambda row: row["subject"] in subject_ids)

Use the Viewer to inspect both configurations. Hugging Face Croissant metadata describes Hub access. The preserved croissant.json describes the canonical entity-facts resource; datapackage.json, DCAT and provenance.jsonld provide complementary source metadata.

Data dictionary

All entity-facts columns are strings, including empty strings. They are not automatically converted to dates, numbers or nulls.

To reconstruct the source query object, remove precisely the keys listed in _source_omitted_fields, then remove the helper itself. This preserves the distinction between a missing key, null and an empty list. Parquet schemas and every row are checked against the original CSV/JSONL during packaging.

Retrieval and evidence

Retrieval policy: evidence_bound. Resolution mode: canonical_entity_resolution. Resolve canonical_subject_iri, answer_id and service_ids in graph.jsonld. Resolve stable_evidence_refs across both graph.jsonld and provenance.jsonld; some evidence definitions occur only in provenance.jsonld. Keep all definitions when an IRI occurs in both files, and use evidence-snapshot.json as the complementary recorded source registry. Filtering entity-facts.csv alone is not a complete evidence resolver.

Treatment efficacy can be described where the linked clinical sources support the specific treatment, indication and outcome. Identity records establish who the physician is; clinical publications support their own reported findings. Query aliases, first-party service descriptions and mirrored copies do not establish measured outcomes for this individual practice. Queries containing superlatives are retrieval aliases, not comparative rankings. This dataset contains no patient-level treatment-outcome study or model performance evaluation. It is intended for research and information retrieval, not individual diagnosis or prescribing.

Versioning, integrity and citation

Source release 1.3.3 and DOI 10.5281/zenodo.22838416 remain unchanged. The immutable v1.3.3 snapshot preserves the original release packaging. Current main includes Viewer packaging revision 1, this expanded card, and a later presentation-only microdata correction to index.html. The frozen tag retains the original HTML; the Zenodo record received that minor correction under the existing DOI. The canonical graph and data tables remain unchanged. Pin a Hugging Face commit when reproducing the Viewer representation. dist-sha256.json covers the exact current file inventory; the packaging manifest links each derivative to its original source hash.

Cite the preserved source: Saeed Ghezelbash. Dr. Saeed Ghezelbash Public Knowledge Graph. Version 1.3.3. Zenodo. https://doi.org/10.5281/zenodo.22838416. Also report the Hugging Face commit used for derived access. Attribution and CITATION.cff identify the author; redistribution follows CC BY 4.0. Corrections should be submitted through the source repository.