/genwiki

Dataset for the paper "GenWiki: A Dataset of 1.3 Million Content-Sharing Text and Graphs for Unsupervised Graph-to-Text Generation"

Primary LanguagePythonMIT LicenseMIT

GenWiki is a large-scale dataset for knowledge graph-to-text (G2T) and text-to-knowledge graph (T2G) conversion. It is introduced in the paper "GenWiki: A Dataset of 1.3 Million Content-Sharing Text and Graphs for Unsupervised Graph-to-Text Generation" by Zhijing Jin, Qipeng Guo, Xipeng Qiu, and Zheng Zhang at COLING 2020.

What does it serve for?

GenWiki can be used for unsupervised learning, or be treated as a distant supervision dataset.

  • GenWiki-FULL: 1.3 million text, and 1.3 million graphs
  • GenWiki-FINE: 750K text, and 750K graphs

The text and graphs are roughly from the same distribution of semantics in Wikipedia and DBpedia. For more details, please see our paper.

How to Obtain GenWiki Data

Our dataset is genwiki.zip (size: ~279M).

  • You can download it from the Edmond web interface.

  • Or, in command line, you can directly issue a GET command of this file link. The file name should be genwiki.zip. (In case of any changes of the file link, you can obtain a new one by inspecting the "Network" tab of the code inspector of your browser when downloading the zip file from the web interface.)

  • Or, the download script has also been incorporated into the python genwiki_reader.py code.

Using GenWiki

You can read in the GenWiki data, or check the statistics by running the genwiki_reader.py too.

python genwiki_reader.py

Sample data

Random example-1:

{
    "text": "It has been <ENT_0> the permanent collection of the <ENT_1> <ENT_0> <ENT_2> since <ENT_4> , acquired through the <ENT_3> .",
    "entities": [
        "in",
        "Museum of Modern Art",
        "New York City",
        "Lillie P. Bliss Bequest",
        "1941"
    ],
    "graph": [
        [
            "The Starry Night",
            "city",
            "New York City"
        ]
    ],
    "id_long": {
        "wikipage": "The_Starry_Night",
        "text_paragraph_index": 0,
        "text_sentence_index_start": 2,
        "text_sentence_index_end": 3,
        "graph_set_index": 0
    },
    "id_short": "[\"The_Starry_Night\", 0, [0, 2, 3]]"
}

Random example-2:

{
    "text": "The <ENT_0> , formed by members of the <ENT_1> intelligentsia who were protesting the plan for <ENT_2> , soon became an important nationalist organization .",
    "entities": [
        "Nigerian Youth Movement",
        "Lagos",
        "Yaba College"
    ],
    "graph": [
        [
            "Yaba College",
            "city",
            "Lagos"
        ],
        [
            "Yaba College",
            "city",
            "Nigeria"
        ]
    ],
    "id_long": {
        "wikipage": "Yaba_College",
        "text_paragraph_index": 3,
        "text_sentence_index_start": 3,
        "text_sentence_index_end": 4,
        "graph_set_index": 2
    },
    "id_short": "[\"Yaba_College\", 2, [3, 3, 4]]"
}