---
license: mit
---

# GQA: Graph Question Answering

This dataset is asks models to make use of embedded graph for question answering.

Stats:
- train: 57,043
- test: 2,890

An exmaple of the dataset is as follows:

```json
{
  "id": "mcwq-176119",
  "question": "What was executive produced by Scott Spiegel , Boaz Yakin , and Quentin Tarantino , executive produced by My Best Friend's Birthday 's editor and star , and edited by George Folsey",
  "answers": [
    "Hostel: Part II"
  ],
  "subgraph": {
    "entities": [
      "Q1401104",
      "Q887636",
      "Q1048645",
      "Q3772",
      "Q965826"
    ],
    "relations": [
      "P1431",
      "P1040"
    ],
    "adjacency": [[2, 1, 0],
      [2, 0, 3],
      [2, 0, 1],
      [2, 0, 4]
    ],
    "entity_labels": [
      "george folsey, jr.",
      "boaz yakin",
      "hostel: part ii",
      "quentin jerome tarantino",
      "scott spiegel"
    ],
    "relation_labels": [
      "showrunner",
      "film editor"
    ]
  },
  "sparql": "SELECT DISTINCT ?x0 WHERE {\n?x0 wdt:P1040 wd:Q1401104 .\n?x0 wdt:P1431 ?x1 .\n?x0 wdt:P1431 wd:Q3772 .\n?x0 wdt:P1431 wd:Q887636 .\n?x0 wdt:P1431 wd:Q965826 .\nwd:Q1480733 wdt:P161 ?x1 .\nwd:Q1480733 wdt:P1040 ?x1\n}"
}
```