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

An Analysis Toolkit for Natural Language Generation (Translation, Captioning, Summarization, etc.)

☀️ Measuring the accuracy of BBC weather forecasts in Honolulu, USA

A PyTorch-based model pruning toolkit for pre-trained language models

Application to help find best train itinerary, uses speech to text, has a spam filter to segregate invalid inputs, NLP and Pathfinding algos.

Python package for performing Entity and Text Matching using Deep Learning.

Modular and extensible speech recognition library leveraging pytorch-lightning and hydra.

Python3 to Crystal Translation using Python AST Walker

New Modeling The Background CodeBase

This repository contains the code for "Generating Datasets with Pretrained Language Models".

Python generation script for BitBirds

It analyze the sentiment of the user, whether it is postive or negative.

Graphical user interface for Argos Translate

NeuralQA: A Usable Library for Question Answering on Large Datasets with BERT

An easier way to build neural search on the cloud

Creating an LSTM model to generate music

Tensorflow implementation of paper: Learning to Diagnose with LSTM Recurrent Neural Networks.

Neural network sequence labeling model

LV-BERT: Exploiting Layer Variety for BERT (Findings of ACL 2021)

Source code for the paper "TearingNet: Point Cloud Autoencoder to Learn Topology-Friendly Representations"

BiQE: Code and dataset for the BiQE paper