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

Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context

Simple text to phones converter for multiple languages

Chinese version of GPT2 training code, using BERT tokenizer.

A framework for training and evaluating AI models on a variety of openly available dialogue datasets.

Intent parsing and slot filling in PyTorch with seq2seq + attention

Research Code for NeurIPS 2020 Spotlight paper "Large-Scale Adversarial Training for Vision-and-Language Representation Learning": UNITER adversarial training part

Transformer - A TensorFlow Implementation of the Transformer: Attention Is All You Need

Chinese segmentation library

This repository details the steps in creating a Part of Speech tagger using Trigram Hidden Markov Models and the Viterbi Algorithm without using external libraries.

ZUNIT - Toward Zero-Shot Unsupervised Image-to-Image Translation

Pretty-doc - Composable text objects with python

A Python 3.6+ package to run .many files, where many programs written in many languages may exist in one file.

NLP applications using deep learning.

The code for the Subformer, from the EMNLP 2021 Findings paper: "Subformer: Exploring Weight Sharing for Parameter Efficiency in Generative Transformers", by Machel Reid, Edison Marrese-Taylor, and Yutaka Matsuo

A natural language modeling framework based on PyTorch

Grading tools for Advanced NLP (11-711)Grading tools for Advanced NLP (11-711)

"Investigating the Limitations of Transformers with Simple Arithmetic Tasks", 2021

A modular framework for vision & language multimodal research from Facebook AI Research (FAIR)

Athena is an open-source implementation of end-to-end speech processing engine.

Traditional Chinese Text Recognition Dataset: Synthetic Dataset and Labeled Data