Skip to content

Latest commit

 

History

History
122 lines (92 loc) · 4.31 KB

README.md

File metadata and controls

122 lines (92 loc) · 4.31 KB

English | 简体中文

CodeKGC-Code Language Models for Knowledge Graph Construction

To better address Relational Triple Extraction (rte) task in Knowledge Graph Construction, we have designed code-style prompts to model the structure of Relational Triple, and used Code-LLMs to generate more accurate predictions. The key step of code-style prompt construction is to transform (text, output triples) pairs into semantically equivalent program language written in Python.

Data and Configuration

  • Data

    The example data of conll04 dataset is stored in the codekgc/data folder. The entire data is available in here. Users can customize your own data, but it is necessary to follow the given data format.

  • Configuration

    The codekgc/config.json file contains the set parameters. The parameters required to load files and call the openai models are passed in through this file.

    Descriptions of these parameters are as follows:

    • schema_path defines the file path of the schema prompt. Schema prompt contains the pre-defined Python classes including Relation class, Entity class, Triple class and Extract class.

      The data format of schema prompt is as follows:

      from typing import List
class Rel:
    def __init__(self, name: str):
        self.name = name
class Work_for(Rel):
...
class Entity:
    def __init__(self, name: str):
        self.name = name
class person(Entity):
...
class Triple:
    def __init__(self, head: Entity, relation: Rel, tail: Entity):
        self.head = head
        self.relation = relation
        self.tail = tail
class Extract:
    def __init__(self, triples: List[Triple] = []):
        self.triples = triples

    • ICL_path defines the file path of in-context examples.

      The data format of ICL prompt is as follows:

      """ In 1856 , the 28th President of the United States , Thomas Woodrow Wilson , was born in Staunton , Va . """
extract = Extract([Triple(person('Thomas Woodrow Wilson'), Rel('Live in'), location('Staunton , Va')),])
...

    • example_path defines the file path of the test example in conll04 dataset.

    • openai_key is your api key of openai.

    • engine, temperature, max_tokens, n... are the parameters required to pass in to call the openai api.

Run and Examples

Once the parameters are set, you can directly run the codekgc.py：

>> cd codekgc
>> python codekgc.py

Below are input and output examples for Relational Triple Extraction (rte) task using code-style prompts:

Input:

from typing import List
class Rel:
...(schema prompt)

""" In 1856 , the 28th President..."""
extract = Extract([Triple(person('Thomas Woodrow Wilson'), Rel('Live in'), location('Staunton , Va')),])
...(in-context examples)

""" Boston University 's Michael D. Papagiannis said he believes the crater was created 100 million years ago when a 50-mile-wide meteorite slammed into the Earth . """

Output:

extract = Extract([Triple(person('Michael D. Papagiannis'), Rel('Work for'), organization('Boston University')),])

Citation

If you use the code, please cite the following paper:

@article{DBLP:journals/corr/abs-2304-09048,
  author       = {Zhen Bi and
                  Jing Chen and
                  Yinuo Jiang and
                  Feiyu Xiong and
                  Wei Guo and
                  Huajun Chen and
                  Ningyu Zhang},
  title        = {CodeKGC: Code Language Model for Generative Knowledge Graph Construction},
  journal      = {CoRR},
  volume       = {abs/2304.09048},
  year         = {2023},
  url          = {https://doi.org/10.48550/arXiv.2304.09048},
  doi          = {10.48550/arXiv.2304.09048},
  eprinttype    = {arXiv},
  eprint       = {2304.09048},
  timestamp    = {Mon, 24 Apr 2023 15:03:18 +0200},
  biburl       = {https://dblp.org/rec/journals/corr/abs-2304-09048.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}