• Journal of Information Processing Systems
• /
• v.19 no.5
• /
• pp.641-651
• /
• 2023

Malicious hate speech and gender bias comments are common in online communities, causing social problems in our society. Gender bias and hate speech detection has been investigated. However, it is difficult because there are diverse ways to express them in words. To solve this problem, we attempted to detect malicious comments in a Korean hate speech dataset constructed in 2020. We explored bidirectional encoder representations from transformers (BERT)-based deep learning models utilizing hyperparameter tuning, data sampling, and logits ensembles with a label distribution. We evaluated our model in Kaggle competitions for gender bias, general bias, and hate speech detection. For gender bias detection, an F1-score of 0.7711 was achieved using an ensemble of the Soongsil-BERT and KcELECTRA models. The general bias task included the gender bias task, and the ensemble model achieved the best F1-score of 0.7166.

• International Journal of Computer Science & Network Security
• /
• v.23 no.1
• /
• pp.112-119
• /
• 2023

Traditionally used for networking computers and communications, the Internet has been evolving from the beginning. Internet is the backbone for many things on the web including social media. The concept of social networking which started in the early 1990s has also been growing with the internet. Social Networking Sites (SNSs) sprung and stayed back to an important element of internet usage mainly due to the services or provisions they allow on the web. Twitter and Facebook have become the primary means by which most individuals keep in touch with others and carry on substantive conversations. These sites allow the posting of photos, videos and support audio and video storage on the sites which can be shared amongst users. Although an attractive option, these provisions have also culminated in issues for these sites like posting offensive material. Though not always, users of SNSs have their share in promoting hate by their words or speeches which is difficult to be curtailed after being uploaded in the media. Hence, this article outlines a process for extracting user reviews from the Twitter corpus in order to identify instances of hate speech. Through the use of MPCA (Modified Principal Component Analysis) and ECNN, we are able to identify instances of hate speech in the text (Enhanced Convolutional Neural Network). With the use of NLP, a fully autonomous system for assessing syntax and meaning can be established (NLP). There is a strong emphasis on pre-processing, feature extraction, and classification. Cleansing the text by removing extra spaces, punctuation, and stop words is what normalization is all about. In the process of extracting features, these features that have already been processed are used. During the feature extraction process, the MPCA algorithm is used. It takes a set of related features and pulls out the ones that tell us the most about the dataset we give itThe proposed categorization method is then put forth as a means of detecting instances of hate speech or abusive language. It is argued that ECNN is superior to other methods for identifying hateful content online. It can take in massive amounts of data and quickly return accurate results, especially for larger datasets. As a result, the proposed MPCA+ECNN algorithm improves not only the F-measure values, but also the accuracy, precision, and recall.

• Annual Conference of KIPS
• /
• 2021.11a
• /
• pp.606-609
• /
• 2021

최근 한국에서도 이루다의 윤리 이슈를 기점으로 딥러닝 모델의 윤리적 언어학습 필요성이 대두되었다. 그럼에도 불구하고 영어 데이터에 비해 한국어 데이터는 Korean Hate Speech Detection Dataset 이 유일하다. 이번 연구에서는 기존 데이터 세트의 유연성이 떨어지고 세부 라벨이 제한적이라는 문제를 개선한 새로운 데이터 세트를 제안하고, 해당 데이터 세트에 대하여 다양한 신경망 분류 모델을 적용한 벤치마크 결과를 공개한다.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.12
• /
• pp.1595-1603
• /
• 2020

Online defamation incidents such as Internet news comments on portal sites, SNS, and community sites are increasing in recent years. Bias and hate expressions threaten online service users in various forms, such as invasion of privacy and personal attacks, and defamation issues. In the past few years, academia and industry have been approaching in various ways to solve this problem The purpose of this study is to build a dataset and experiment with deep learning classification modeling for detecting various bias expressions as well as hate expressions. The dataset was annotated 7 labels that 10 personnel cross-checked. In this study, each of the 7 classes in a dataset of about 137,111 Korean internet news comments is binary classified and analyzed through deep learning techniques. The Proposed technique used in this study is multi-channel CNN model with attention. As a result of the experiment, the weighted average f1 score was 70.32% of performance.

• Journal of Web Engineering
• /
• v.21 no.3
• /
• pp.677-692
• /
• 2022

Recognizing the offensive, abusive, and profanity of multimedia content on the web has been a challenge to keep the web environment for user's freedom of speech. As profanity filtering function has been developed and applied in text, audio, and video context in platforms such as social media, entertainment, and education, the number of methods to trick the web-based application also has been increased and became a new issue to be solved. Compared to commonly developed toxic content detection systems that use lexicon and keyword-based detection, this work tries to embrace a different approach by the meaning of the sentence. Meaning representation is a way to grasp the meaning of linguistic input. This work proposed a data-driven approach utilizing Abstract meaning Representation to extract the meaning of the online text content into a convolutional neural network to detect level profanity. This work implements the proposed model in two kinds of datasets from the Offensive Language Identification Dataset and other datasets from the Offensive Hate dataset merged with the Twitter Sentiment Analysis dataset. The results indicate that the proposed model performs effectively, and can achieve a satisfactory accuracy in recognizing the level of online text content toxicity.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.518-523
• /
• 2021

챗봇과 같은 대화형 에이전트 사용이 증가하면서 채팅에서의 혐오 표현 사용도 더불어 증가하고 있다. 혐오 표현을 자동으로 탐지하려는 노력은 다양하게 시도되어 왔으나, 챗봇 데이터를 대상으로 한 혐오 표현 탐지 연구는 여전히 부족한 실정이다. 이 연구는 혐오 표현을 포함한 챗봇-사용자 대화 데이터 35만 개에 한국어 말뭉치로 학습된 KoELETRA 기반 혐오 탐지 모델을 적용하여, 챗봇-사람 데이터셋에서의 혐오 표현 탐지의 성능과 한계점을 검토하였다. KoELECTRA 혐오 표현 분류 모델은 챗봇 데이터셋에 대해 가중 평균 F1-score 0.66의 성능을 보였으며, 오탈자에 대한 취약성, 맥락 미반영으로 인한 편향 강화, 가용한 데이터의 정확도 문제가 주요한 한계로 포착되었다. 이 연구에서는 실험 결과에 기반해 성능 향상을 위한 방향성을 제시한다.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.577-580
• /
• 2023

악성댓글은 인터넷상에서 정서적, 심리적 피해를 주는 문제로 인식되어 왔다. 본 연구는 한국어 악성댓글 탐지 분석을 위해 KcBERT 및 다양한 모델을 활용하여 성능을 비교하였다. 또한, 공개된 한국어 악성댓글 데이터가 부족한 것을 해소하기 위해 기계 번역을 이용하고, 다국어 언어 모델(Multilingual Model) mBERT를 활용하였다. 다양한 실험을 통해 KcBERT를 미세 조정한 모델의 정확도 및 F1-score가 타 모델에 비해 의미 있는 결과임을 확인할 수 있었다.

• The Transactions of the Korea Information Processing Society
• /
• v.13 no.4
• /
• pp.181-188
• /
• 2024

Traditional profanity detection methods have limitations in identifying intentionally altered profanities. This paper introduces a new method based on Named Entity Recognition, a subfield of Natural Language Processing. We developed a profanity detection technique using sequence labeling, for which we constructed a dataset by labeling some profanities in Korean malicious comments and conducted experiments. Additionally, to enhance the model's performance, we augmented the dataset by labeling parts of a Korean hate speech dataset using one of the large language models, ChatGPT, and conducted training. During this process, we confirmed that filtering the dataset created by the large language model by humans alone could improve performance. This suggests that human oversight is still necessary in the dataset augmentation process.