CGRE Open-source Chinese Relation Extraction Model - Efficiently Extract Entity Relationships with Generative Methods

CGRE CNDBPedia Generative Relation Extraction

Developed by fanxiao

CGRE is a Chinese end-to-end relation extraction model based on BART, employing a generative approach to achieve entity relation extraction, demonstrating excellent performance on multiple Chinese relation extraction datasets.

Knowledge Graph

Transformers

#Chinese Relation Extraction #Generative Models #Encyclopedia Knowledge Base

Downloads 63

Release Time : 6/17/2022

Model Overview

CGRE is a generative entity relation extraction model based on BART as the backbone network, capable of directly generating linearized triplet relations from sentences.

Model Features

Generative Relation Extraction

Uses a generative approach to directly output triplets, simplifying the traditional relation extraction process.

Chinese Optimization

Specifically optimized for Chinese text, achieving excellent performance on multiple Chinese relation extraction datasets.

End-to-End Model

No complex preprocessing or postprocessing required, enabling end-to-end relation extraction.

Distant Supervision Training

Trained using distantly supervised data from Chinese encyclopedias to enhance model generalization.

Model Capabilities

Entity Relation Extraction

Chinese Text Understanding

Triplet Generation

Use Cases

Information Extraction

Encyclopedia Knowledge Extraction

Extract entity relations from encyclopedia-like texts

Achieves industry-leading performance on datasets like DuIE

Business Intelligence Analysis

Extract company-product relations from business news

🚀 CGRE - Generation-based Relation Extraction Model

CGRE is a state-of-the-art (SOTA) Chinese end-to-end relation extraction model. It leverages the BART architecture as its backbone, offering high performance and ease of use in various Chinese relation extraction tasks.

✨ Features

SOTA Performance: It stands as a SOTA Chinese end-to-end relation extraction model, using BART as the backbone.
Data and Pretraining: Trained on Distant-supervised data from cndbpedia and pretrained from the checkpoint of fnlp/bart-base-chinese.
High Adaptability: Achieves SOTA results on many Chinese relation extraction datasets, including DuIE~1.0, DuIE~2.0, HacRED, etc.
User-Friendly: Simple to use, similar to a normal generation task. The input is a sentence, and the output is linearized triples. For example, given the input "姚明是一名NBA篮球运动员" (Yao Ming is an NBA basketball player), the output is "[subj]姚明[obj]NBA[rel]公司[obj]篮球运动员[rel]职业" ([subj]Yao Ming[obj]NBA[rel]Company[obj]Basketball Player[rel]Occupation).

📦 Installation

The installation mainly involves setting up the necessary Python libraries. You need to have transformers installed. You can install it using pip install transformers.

💻 Usage Examples

Basic Usage

from transformers import BertTokenizer, BartForConditionalGeneration

model_name = 'fnlp/bart-base-chinese'

tokenizer_kwargs = {
    "use_fast": True,
    "additional_special_tokens": ['<rel>', '<obj>', '<subj>'],
} # if cannot see tokens in model card please open readme file

tokenizer = BertTokenizer.from_pretrained(model_name, **tokenizer_kwargs)

model = BartForConditionalGeneration.from_pretrained('./CGRE_CNDBPedia-Generative-Relation-Extraction')

sent = "姚明是一名NBA篮球运动员"
max_source_length = 128
max_target_length = 128
inputs = tokenizer(sent, max_length=max_source_length, padding="max_length", truncation=True, return_tensors="pt")

params = {"decoder_start_token_id":0,"early_stopping":False,"no_repeat_ngram_size":0,"length_penalty": 0,"num_beams":20,"use_cache":True}

out_id = model.generate(inputs["input_ids"], attention_mask = inputs["attention_mask"], max_length=max_target_length, **params)

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご