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

This is a Prototype of an Ai ChatBot "Tea and Coffee Supplier" using python.

Open source annotation tool for machine learning practitioners.

KoBERT - Korean BERT pre-trained cased (KoBERT)

Code for EmBERT, a transformer model for embodied, language-guided visual task completion.

ALIbaba's Collection of Encoder-decoders from MinD (Machine IntelligeNce of Damo) Lab

Persian-lexicon - A lexicon of 70K unique Persian (Farsi) words

Negative sampling for solving the unlabeled entity problem in NER. ICLR-2021 paper: Empirical Analysis of Unlabeled Entity Problem in Named Entity Recognition.

LewusBot - Twitch ChatBot built in python with twitchio library

Revisiting Pre-trained Models for Chinese Natural Language Processing (Findings of EMNLP 2020)

This is a simple item2vec implementation using gensim for recbole

Course project of [email protected]

Transformer training code for sequential tasks

An easier way to build neural search on the cloud

[Preprint] Escaping the Big Data Paradigm with Compact Transformers, 2021

Spacy-ginza-ner-webapi - Named Entity Recognition API with spaCy and GiNZA

Python utility library for compositing PDF documents with reportlab.

Skipgram Negative Sampling in PyTorch

Library for Russian imprecise rhymes generation

NLP techniques such as named entity recognition, sentiment analysis, topic modeling, text classification with Python to predict sentiment and rating of drug from user reviews.

Code to use Augmented Shapiro Wilks Stopping, as well as code for the paper "Statistically Signifigant Stopping of Neural Network Training"