arXiv subjects

Haewoon Kwak

Publications and source records attributed to Haewoon Kwak.

At least 37 records · Page 2Linked to original sources

arXiv2024

ToBlend: Token-Level Blending With an Ensemble of LLMs to Attack AI-Generated Text Detection

The robustness of AI-content detection models against sophisticated adversarial strategies, such as paraphrasing or word switching, is a rising concern in natural language generation (NLG) applications. This study proposes ToBlend, a novel token-level ensemble text generation method to challenge the robustness of current AI-content detection approaches by utilizing multiple sets of candidate generative large language models (LLMs). By randomly sampling token(s) from candidate LLMs sets, we find ToBlend significantly drops the performance of most mainstream AI-content detection methods. We evaluate the text quality produced under different ToBlend settings based on annotations from experienced human experts. We proposed a fine-tuned Llama3.1 model to distinguish the ToBlend generated text more accurately. Our findings underscore our proposed text generation approach's great potential in deceiving and improving detection models. Our datasets, codes, and annotations are open-sourced.

cs.CL↗

arXiv2024

Can we trust the evaluation on ChatGPT?

ChatGPT, the first large language model (LLM) with mass adoption, has demonstrated remarkable performance in numerous natural language tasks. Despite its evident usefulness, evaluating ChatGPT's performance in diverse problem domains remains challenging due to the closed nature of the model and its continuous updates via Reinforcement Learning from Human Feedback (RLHF). We highlight the issue of data contamination in ChatGPT evaluations, with a case study of the task of stance detection. We discuss the challenge of preventing data contamination and ensuring fair model evaluation in the age of closed and continuously trained models.

cs.CL↗

arXiv2024

Rematch: Robust and Efficient Matching of Local Knowledge Graphs to Improve Structural and Semantic Similarity

Knowledge graphs play a pivotal role in various applications, such as question-answering and fact-checking. Abstract Meaning Representation (AMR) represents text as knowledge graphs. Evaluating the quality of these graphs involves matching them structurally to each other and semantically to the source text. Existing AMR metrics are inefficient and struggle to capture semantic similarity. We also lack a systematic evaluation benchmark for assessing structural similarity between AMR graphs. To overcome these limitations, we introduce a novel AMR similarity metric, rematch, alongside a new evaluation for structural similarity called RARE. Among state-of-the-art metrics, rematch ranks second in structural similarity; and first in semantic similarity by 1--5 percentage points on the STS-B and SICK-R benchmarks. Rematch is also five times faster than the next most efficient metric.

cs.CL↗

arXiv2024

ChatGPT Rates Natural Language Explanation Quality Like Humans: But on Which Scales?

As AI becomes more integral in our lives, the need for transparency and responsibility grows. While natural language explanations (NLEs) are vital for clarifying the reasoning behind AI decisions, evaluating them through human judgments is complex and resource-intensive due to subjectivity and the need for fine-grained ratings. This study explores the alignment between ChatGPT and human assessments across multiple scales (i.e., binary, ternary, and 7-Likert scale). We sample 300 data instances from three NLE datasets and collect 900 human annotations for both informativeness and clarity scores as the text quality measurement. We further conduct paired comparison experiments under different ranges of subjectivity scores, where the baseline comes from 8,346 human annotations. Our results show that ChatGPT aligns better with humans in more coarse-grained scales. Also, paired comparisons and dynamic prompting (i.e., providing semantically similar examples in the prompt) improve the alignment. This research advances our understanding of large language models' capabilities to assess the text explanation quality in different configurations for responsible AI development.

cs.CL↗

arXiv2024

Benchmarking zero-shot stance detection with FlanT5-XXL: Insights from training data, prompting, and decoding strategies into its near-SoTA performance

We investigate the performance of LLM-based zero-shot stance detection on tweets. Using FlanT5-XXL, an instruction-tuned open-source LLM, with the SemEval 2016 Tasks 6A, 6B, and P-Stance datasets, we study the performance and its variations under different prompts and decoding strategies, as well as the potential biases of the model. We show that the zero-shot approach can match or outperform state-of-the-art benchmarks, including fine-tuned models. We provide various insights into its performance including the sensitivity to instructions and prompts, the decoding strategies, the perplexity of the prompts, and to negations and oppositions present in prompts. Finally, we ensure that the LLM has not been trained on test datasets, and identify a positivity bias which may partially explain the performance differences across decoding strategie

cs.CL↗

arXiv2023

Public Perception of Generative AI on Twitter: An Empirical Study Based on Occupation and Usage

The emergence of generative AI has sparked substantial discussions, with the potential to have profound impacts on society in all aspects. As emerging technologies continue to advance, it is imperative to facilitate their proper integration into society, managing expectations and fear. This paper investigates users' perceptions of generative AI using 3M posts on Twitter from January 2019 to March 2023, especially focusing on their occupation and usage. We find that people across various occupations, not just IT-related ones, show a strong interest in generative AI. The sentiment toward generative AI is generally positive, and remarkably, their sentiments are positively correlated with their exposure to AI. Among occupations, illustrators show exceptionally negative sentiment mainly due to concerns about the unethical usage of artworks in constructing AI. People use ChatGPT in diverse ways, and notably the casual usage in which they "play with" ChatGPT tends to associate with positive sentiments. After the release of ChatGPT, people's interest in AI in general has increased dramatically; however, the topic with the most significant increase and positive sentiment is related to crypto, indicating the hype-worthy characteristics of generative AI. These findings would offer valuable lessons for policymaking on the emergence of new technology and also empirical insights for the considerations of future human-AI symbiosis.

cs.SI↗

arXiv2023

YouNICon: YouTube's CommuNIty of Conspiracy Videos

Conspiracy theories are widely propagated on social media. Among various social media services, YouTube is one of the most influential sources of news and entertainment. This paper seeks to develop a dataset, YOUNICON, to enable researchers to perform conspiracy theory detection as well as classification of videos with conspiracy theories into different topics. YOUNICON is a dataset with a large collection of videos from suspicious channels that were identified to contain conspiracy theories in a previous study (Ledwich and Zaitsev 2020). Overall, YOUNICON will enable researchers to study trends in conspiracy theories and understand how individuals can interact with the conspiracy theory producing community or channel. Our data is available at: https://doi.org/10.5281/zenodo.7466262.

cs.CY↗

arXiv2023

Wearing Masks Implies Refuting Trump?: Towards Target-specific User Stance Prediction across Events in COVID-19 and US Election 2020

People who share similar opinions towards controversial topics could form an echo chamber and may share similar political views toward other topics as well. The existence of such connections, which we call connected behavior, gives researchers a unique opportunity to predict how one would behave for a future event given their past behaviors. In this work, we propose a framework to conduct connected behavior analysis. Neural stance detection models are trained on Twitter data collected on three seemingly independent topics, i.e., wearing a mask, racial equality, and Trump, to detect people's stance, which we consider as their online behavior in each topic-related event. Our results reveal a strong connection between the stances toward the three topical events and demonstrate the power of past behaviors in predicting one's future behavior.

cs.SI↗

arXiv2023

Chain of Explanation: New Prompting Method to Generate Higher Quality Natural Language Explanation for Implicit Hate Speech

Recent studies have exploited advanced generative language models to generate Natural Language Explanations (NLE) for why a certain text could be hateful. We propose the Chain of Explanation (CoE) Prompting method, using the heuristic words and target group, to generate high-quality NLE for implicit hate speech. We improved the BLUE score from 44.0 to 62.3 for NLE generation by providing accurate target information. We then evaluate the quality of generated NLE using various automatic metrics and human annotations of informativeness and clarity scores.

cs.CL↗

arXiv2023

Is ChatGPT better than Human Annotators? Potential and Limitations of ChatGPT in Explaining Implicit Hate Speech

Recent studies have alarmed that many online hate speeches are implicit. With its subtle nature, the explainability of the detection of such hateful speech has been a challenging problem. In this work, we examine whether ChatGPT can be used for providing natural language explanations (NLEs) for implicit hateful speech detection. We design our prompt to elicit concise ChatGPT-generated NLEs and conduct user studies to evaluate their qualities by comparison with human-written NLEs. We discuss the potential and limitations of ChatGPT in the context of implicit hateful speech research.

cs.CL↗

arXiv2023

Political Honeymoon Effect on Social Media: Characterizing Social Media Reaction to the Changes of Prime Minister in Japan

New leaders in democratic countries typically enjoy high approval ratings immediately after taking office. This phenomenon is called the honeymoon effect and is regarded as a significant political phenomenon; however, its mechanism remains underexplored. Therefore, this study examines how social media users respond to changes in political leadership in order to better understand the honeymoon effect in politics. In particular, we constructed a 15-year Twitter dataset on eight change timings of Japanese prime ministers consisting of 6.6M tweets and analyzed them in terms of sentiments, topics, and users. We found that, while not always, social media tend to show a honeymoon effect at the change timings of prime minister. The study also revealed that sentiment about prime ministers differed by topic, indicating that public expectations vary from one prime minister to another. Furthermore, the user base was largely replaced before and after the change in the prime minister, and their sentiment was also significantly different. The implications of this study would be beneficial for administrative management.

cs.SI↗

arXiv2022

Characterizing Spontaneous Ideation Contest on Social Media: Case Study on the Name Change of Facebook to Meta

Collecting good ideas is vital for organizations, especially companies, to retain their competitiveness. Social media is gathering attention as a place to extract ideas efficiently; however, the characteristics of ideas and the posters of ideas on social media are underexamined. Thus, this study aims to characterize spontaneous ideation contests among social media users by taking an event of Facebook's name change to Meta as a case study. As a dataset, we comprehensively collect tweets containing new acronyms of Big Tech companies, which we treat as an "idea" in this work. In the analysis, we especially focus on the diversity of ideas, which would be the main reason for enlisting social media for idea generation. As the main results, we discovered that social media users offered a wider range of ideas than those in mainstream media. The follow-follower network of the users suggested that the users' position on the network is related to the preferred ideas. Additionally, we discovered a link between the amount of user interaction on social media and the diversity of ideas. This study would promote the use of social media as a part of open innovation and co-creation processes in the industry.

cs.SI↗

arXiv2022

"This is Fake News": Characterizing the Spontaneous Debunking from Twitter Users to COVID-19 False Information

False information spreads on social media, and fact-checking is a potential countermeasure. However, there is a severe shortage of fact-checkers; an efficient way to scale fact-checking is desperately needed, especially in pandemics like COVID-19. In this study, we focus on spontaneous debunking by social media users, which has been missed in existing research despite its indicated usefulness for fact-checking and countering false information. Specifically, we characterize the tweets with false information, or fake tweets, that tend to be debunked and Twitter users who often debunk fake tweets. For this analysis, we create a comprehensive dataset of responses to fake tweets, annotate a subset of them, and build a classification model for detecting debunking behaviors. We find that most fake tweets are left undebunked, spontaneous debunking is slower than other forms of responses, and spontaneous debunking exhibits partisanship in political topics. These results provide actionable insights into utilizing spontaneous debunking to scale conventional fact-checking, thereby supplementing existing research from a new perspective.

cs.SI↗

arXiv2022

You Have Earned a Trophy: Characterize In-Game Achievements and Their Completions

Achievement systems have been actively adopted in gaming platforms to maintain players' interests. Among them, trophies in PlayStation games are one of the most successful achievement systems. While the importance of trophy design has been casually discussed in many game developers' forums, there has been no systematic study of the historical dataset of trophies yet. In this work, we construct a complete dataset of PlayStation games and their trophies and investigate them from both the developers' and players' perspectives.

cs.HC↗

arXiv2022

Modeling Political Activism around Gun Debate via Social Media

The United States have some of the highest rates of gun violence among developed countries. Yet, there is a disagreement about the extent to which firearms should be regulated. In this study, we employ social media signals to examine the predictors of offline political activism, at both population and individual level. We show that it is possible to classify the stance of users on the gun issue, especially accurately when network information is available. Alongside socioeconomic variables, network information such as the relative size of the two sides of the debate is also predictive of state-level gun policy. On individual level, we build a statistical model using network, content, and psycho-linguistic features that predicts real-life political action, and explore the most predictive linguistic features. Thus, we argue that, alongside demographics and socioeconomic indicators, social media provides useful signals in the holistic modeling of political engagement around the gun debate.

cs.SI↗

arXiv2022

Who Is Missing? Characterizing the Participation of Different Demographic Groups in a Korean Nationwide Daily Conversation Corpus

A conversation corpus is essential to build interactive AI applications. However, the demographic information of the participants in such corpora is largely underexplored mainly due to the lack of individual data in many corpora. In this work, we analyze a Korean nationwide daily conversation corpus constructed by the National Institute of Korean Language (NIKL) to characterize the participation of different demographic (age and sex) groups in the corpus.

cs.CL↗

arXiv2022

Understanding Toxicity Triggers on Reddit in the Context of Singapore

While the contagious nature of online toxicity sparked increasing interest in its early detection and prevention, most of the literature focuses on the Western world. In this work, we demonstrate that 1) it is possible to detect toxicity triggers in an Asian online community, and 2) toxicity triggers can be strikingly different between Western and Eastern contexts.

cs.CY↗

arXiv2021

Predicting Anti-Asian Hateful Users on Twitter during COVID-19

We investigate predictors of anti-Asian hate among Twitter users throughout COVID-19. With the rise of xenophobia and polarization that has accompanied widespread social media usage in many nations, online hate has become a major social issue, attracting many researchers. Here, we apply natural language processing techniques to characterize social media users who began to post anti-Asian hate messages during COVID-19. We compare two user groups -- those who posted anti-Asian slurs and those who did not -- with respect to a rich set of features measured with data prior to COVID-19 and show that it is possible to predict who later publicly posted anti-Asian slurs. Our analysis of predictive features underlines the potential impact of news media and information sources that report on online hate and calls for further investigation into the role of polarized communication networks and news media.

cs.CY↗