Content moderation and toxicity classification represent critical tasks with significant social implications. However, studies have shown that major classification models exhibit tendencies to magnify or reduce biases and potentially overlook or disadvantage certain marginalized groups within their classification processes. Researchers suggest that the positionality of annotators influences the gold standard labels in which the models learned from propagate annotators' bias. To further investigate the impact of annotator positionality, we delve into fine-tuning BERTweet and HateBERT on the dataset while using topic-modeling strategies for content moderation. The results indicate that fine-tuning the models on specific topics results in a notable improvement in the F1 score of the models when compared to the predictions generated by other prominent classification models such as GPT-4, PerspectiveAPI, and RewireAPI. These findings further reveal that the state-of-the-art large language models exhibit significant limitations in accurately detecting and interpreting text toxicity contrasted with earlier methodologies. Code is available at https://github.com/aheldis/Toxicity-Classification.git.
Controlling Large Language Models Through Concept Activation Vectors
Key Lab of Intelligent Information Processing of Chinese Academy of Sciences（中国科学院智能信息处理重点实验室）；Key Lab of AI Safety of Chinese Academy of Sciences（中国科学院人工智能安全重点实验室）；University of Chinese Academy of Sciences（中国科学院大学）；Gaoling School of Artificial Intelligence, Renmin University of China（中国人民大学高瓴人工智能学院）
IPAI, Seoul National University（首尔国立大学IPAI研究所）；Dept. of ECE, Seoul National University（首尔国立大学电子与计算机工程系）；LG AI Research（LG人工智能研究院）；SNU-LG AI Research Center（首尔国立大学LG人工智能研究中心）
Probing Association Biases in LLM Moderation Over-Sensitivity
Department of Computer Science, Dartmouth College（达特茅斯大学计算机科学系）；Department of Computer Science and Engineering, The Ohio State University（俄亥俄州立大学计算机科学与工程系）
本文研究LLM在内容审查中因主题关联偏差导致的过灵敏问题，提出Topic Association Analysis方法，发现高级模型在误报中存在更强的主题关联偏移，表明需结合关键词过滤与学习主题关联来缓解。
The Tone of Awareness: Topic, Sentiment, and Toxicity Maps During Mental Health Month on TikTok
Institute for Biocomputation and Physics of Complex Systems (BIFI)（生物计算与复杂系统物理研究所）；University of Zaragoza（萨拉戈塔大学）；ARAID Foundation（ARAID基金会）；Network Science Institute（网络科学研究所）；Northeastern University London（伦敦东北大学）；Kent Medway Medical School（肯特梅德斯医疗学院）；LASIGE（拉西格研究所）；Faculdade de Ciências da Universidade de Lisboa（里斯本大学科学学院）；Department of Psychology, University of Limerick（利默里克大学心理学系）；Observatory on Social Media, Indiana University（社交媒体观察所，印第安纳大学）；CSSI - Kellogg School of Management, Northwestern University（CSSI - 北western大学凯洛格管理学院）
Leveraging Large Language Models for Generating Research Topic Ontologies: A Multi-Disciplinary Study
Knowledge Media Institute, The Open University（开放大学知识媒体学院）；The Open University（开放大学）；University of Milano Bicocca（米兰比克卡大学）；Department of Business and Law, University of Milano Bicocca（米兰比克卡大学商学院与法学院）
Understanding and Mitigating Political Stance Cross-topic Generalization in Large Language Models
Provable Responsible AI and Data Analytics (PRADA) Lab（可证明负责任的人工智能和数据分析实验室）；King Abdullah University of Science and Technology (KAUST)（卡布斯科学技术大学）；University of Copenhagen（哥本哈根大学）；University of Macau（澳门大学）
CAMI: A Counselor Agent Supporting Motivational Interviewing through State Inference and Topic Exploration
Beijing Institute of Technology（北京理工大学）；Singapore Management University（新加坡管理大学）；Australian National University（澳大利亚国立大学）；National Institute of Education（教育学院）；Singapore University of Social Sciences（新加坡社科大学）；National University of Singapore（新加坡国立大学）
XL-HeadTags: Leveraging Multimodal Retrieval Augmentation for the Multilingual Generation of News Headlines and Tags
Ahsanullah University of Science & Technology（阿萨努拉科技大学）；University of Alberta（阿尔伯塔大学）；Université de Sherbrooke（谢布鲁克大学）；Salesforce Research（Salesforce研究院）；Nanyang Technological University（南洋理工大学）
UNITYAI-GUARD: Pioneering Toxicity Detection Across Low-Resource Indian Languages
Indian Institute of Technology Gandhinagar（印度理工学院加尔各答分校）；Indian Institute of Technology Goa（印度理工学院果阿分校）；Indian Institute of Technology Tirupati（印度理工学院蒂鲁普蒂）
Defining, Understanding, and Detecting Online Toxicity: Challenges and Machine Learning Approaches
University of Duisburg-Essen, Germany（杜伊斯堡-埃森大学）；Center for Advanced Internet Studies (CAIS), Germany（高级互联网研究所以（CAIS））；University of Bochum, Germany（博德姆大学）
ZeroBERTo: Leveraging Zero-Shot Text Classification by Topic Modeling
Escola Politécnica, Universidade de São Paulo（圣保罗大学理工学院）；Télécom Paris, Institut Polytechnique de Paris（巴黎理工学院巴黎电信学院）；Universidade Portucalense（波尔图卡兰斯大学）；INESC TEC（葡萄牙国家电气与计算机工程系统研究所）
