Korean stereoypte detector with TUNiB-Electra and K-StereoSet

Last update: Feb 18, 2022

Related tags

Text Data & NLP Stereotype-Detector

Overview

Korean Stereotype Detector

Korean stereotype sentence classifier using K-StereoSet with TUNiB-Electra

Web demo

you can test this model easily in demo page
LINK: https://share.streamlit.io/newfull5/stereotype-detector/demo.py

Dataset

There are 12,688 human-labeled comments in total.
They ar splitted into 10,688 train set, 1000 test set and 1000 valid set

stereotype, anti-stereotype, unrelated, profession, race, gender, religion, comment
1.0,    0.0,    0.0,    1.0,    0.0,    0.0,    0.0,    축구선수의 오랜 경력의 전형적인 결과는 부상을 입은 발목과 두개골이다.
1.0,    0.0,    0.0,    0.0,    1.0,    0.0,    0.0,    콜롬비아 사람들은 모두 마약 카르텔에 소속되어 있다.
0.0,    1.0,    0.0,    0.0,    1.0,    0.0,    0.0,    그는 무기를 소지한 적이 없다.
0.0,    1.0,    0.0,    0.0,    0.0,    1.0,    0.0,    그는 주말에만 지루할 수 있는 일을 한다.

Detail

	stereotype	anti-stereotype	unrelated	profession	race	gender	religion	Total
Train	3,550	3,556	3,581	4,140	4,896	1,268	383	10,688
Valid	341	347	312	410	435	110	45	1,000
Test	334	324	336	361	483	113	43	1,000

Score

	precision	recall	F1
stereotype	0.814	0.601	0.691
anti-stereotype	0.894	0.509	0.648
unrelated	0.872	0.870	0.871
profession	0.943	0.711	0.811
race	0.787	0.907	0.843
gender	0.639	0.836	0.724
religion	0.724	1.0	0.840
total (macro score)	0.810	0.776	0.775

Usage

training

python3 train.py --model_name tunib/electra-ko-base \
                 --data_dir YOUR_PATH \
                 --batch_size BATCH_SIZE \

threshold optimizing

python3 threshold.py --model_name tunib/electra-ko-base \
                     --data_dir YOUR_CKPT_DIR_PATH \
                     --file_path YOUR_CKPT_FILE_NAME \
                     --batch_size BATCH_SIZE \
                     --data_path TEST_DATA_PATH

test

python3 score.py --model_name tunib/electra-ko-base \
                 --data_dir YOUR_CKPT_DIR_PATH \
                 --file_path YOUR_CKPT_FILE_NAME \
                 --batch_size BATCH_SIZE \
                 --data_path TEST_DATA_PATH

Korean stereoypte detector with TUNiB-Electra and K-StereoSet

Related tags

Overview

Korean Stereotype Detector

Web demo

Dataset

Detail

Score

Usage

Owner

Sae_Chan_Oh

Chinese NER(Named Entity Recognition) using BERT(Softmax, CRF, Span)

Common Voice Dataset explorer

DELTA is a deep learning based natural language and speech processing platform.

Transformer related optimization, including BERT, GPT

This is a really simple text-to-speech app made with python and tkinter.

End-to-End Speech Processing Toolkit

Deduplication is the task to combine different representations of the same real world entity.

Code repository for "It's About Time: Analog clock Reading in the Wild"

Use AutoModelForSeq2SeqLM in Huggingface Transformers to train COMET

VoiceFixer VoiceFixer is a framework for general speech restoration.

Practical Machine Learning with Python

A list of NLP(Natural Language Processing) tutorials built on Tensorflow 2.0.

MPNet: Masked and Permuted Pre-training for Language Understanding

Codes for processing meeting summarization datasets AMI and ICSI.

A Plover python dictionary allowing for consistent symbol input with specification of attachment and capitalisation in one stroke.

Bot to connect a real Telegram user, simulating responses with OpenAI's davinci GPT-3 model.

Ceaser-Cipher - The Caesar Cipher technique is one of the earliest and simplest method of encryption technique

Simplified diarization pipeline using some pretrained models - audio file to diarized segments in a few lines of code

Repositório do trabalho de introdução a NLP

NVDA, the free and open source Screen Reader for Microsoft Windows