Dongwon Lee

Nafis Irtiza Tripto, Adaku Uchendu, Thai Le et al.

Authorship Analysis, also known as stylometry, has been an essential aspect of Natural Language Processing (NLP) for a long time. Likewise, the recent advancement of Large Language Models (LLMs) has made authorship analysis increasingly crucial for distinguishing between human-written and AI-generated texts. However, these authorship analysis tasks have primarily been focused on written texts, not considering spoken texts. Thus, we introduce the largest benchmark for spoken texts - HANSEN (Human ANd ai Spoken tExt beNchmark). HANSEN encompasses meticulous curation of existing speech datasets accompanied by transcripts, alongside the creation of novel AI-generated spoken text datasets. Together, it comprises 17 human datasets, and AI-generated spoken texts created using 3 prominent LLMs: ChatGPT, PaLM2, and Vicuna13B. To evaluate and demonstrate the utility of HANSEN, we perform Authorship Attribution (AA) & Author Verification (AV) on human-spoken datasets and conducted Human vs. AI spoken text detection using state-of-the-art (SOTA) models. While SOTA methods, such as, character ngram or Transformer-based model, exhibit similar AA & AV performance in human-spoken datasets compared to written ones, there is much room for improvement in AI-generated spoken text detection. The HANSEN benchmark is available at: https://huggingface.co/datasets/HANSEN-REPO/HANSEN.

Jooyoung Lee, Thai Le, Jinghui Chen et al.

Past literature has illustrated that language models (LMs) often memorize parts of training instances and reproduce them in natural language generation (NLG) processes. However, it is unclear to what extent LMs "reuse" a training corpus. For instance, models can generate paraphrased sentences that are contextually similar to training samples. In this work, therefore, we study three types of plagiarism (i.e., verbatim, paraphrase, and idea) among GPT-2 generated texts, in comparison to its training data, and further analyze the plagiarism patterns of fine-tuned LMs with domain-specific corpora which are extensively used in practice. Our results suggest that (1) three types of plagiarism widely exist in LMs beyond memorization, (2) both size and decoding methods of LMs are strongly associated with the degrees of plagiarism they exhibit, and (3) fine-tuned LMs' plagiarism patterns vary based on their corpus similarity and homogeneity. Given that a majority of LMs' training data is scraped from the Web without informing content owners, their reiteration of words, phrases, and even core ideas from training sets into generated texts has ethical implications. Their patterns are likely to exacerbate as both the size of LMs and their training data increase, raising concerns about indiscriminately pursuing larger models with larger training corpora. Plagiarized content can also contain individuals' personal and sensitive information. These findings overall cast doubt on the practicality of current LMs in mission-critical writing tasks and urge more discussions around the observed phenomena. Data and source code are available at https://github.com/Brit7777/LM-plagiarism.

Adaku Uchendu, Jooyoung Lee, Hua Shen et al.

Advances in Large Language Models (e.g., GPT-4, LLaMA) have improved the generation of coherent sentences resembling human writing on a large scale, resulting in the creation of so-called deepfake texts. However, this progress poses security and privacy concerns, necessitating effective solutions for distinguishing deepfake texts from human-written ones. Although prior works studied humans' ability to detect deepfake texts, none has examined whether "collaboration" among humans improves the detection of deepfake texts. In this study, to address this gap of understanding on deepfake texts, we conducted experiments with two groups: (1) nonexpert individuals from the AMT platform and (2) writing experts from the Upwork platform. The results demonstrate that collaboration among humans can potentially improve the detection of deepfake texts for both groups, increasing detection accuracies by 6.36% for non-experts and 12.76% for experts, respectively, compared to individuals' detection accuracies. We further analyze the explanations that humans used for detecting a piece of text as deepfake text, and find that the strongest indicator of deepfake texts is their lack of coherence and consistency. Our study provides useful insights for future tools and framework designs to facilitate the collaborative human detection of deepfake texts. The experiment datasets and AMT implementations are available at: https://github.com/huashen218/llm-deepfake-human-study.git

Yeon-Chang Lee, JaeHyun Lee, Michiharu Yamashita et al. · gatech

The problem of career trajectory prediction (CTP) aims to predict one's future employer or job position. While several CTP methods have been developed for this problem, we posit that none of these methods (1) jointly considers the mutual ternary dependency between three key units (i.e., user, position, and company) of a career and (2) captures the characteristic shifts of key units in career over time, leading to an inaccurate understanding of the job movement patterns in the labor market. To address the above challenges, we propose a novel solution, named as CAPER, that solves the challenges via sophisticated temporal knowledge graph (TKG) modeling. It enables the utilization of a graph-structured knowledge base with rich expressiveness, effectively preserving the changes in job movement patterns. Furthermore, we devise an extrapolated career reasoning task on TKG for a realistic evaluation. The experiments on a real-world career trajectory dataset demonstrate that CAPER consistently and significantly outperforms four baselines, two recent TKG reasoning methods, and five state-of-the-art CTP methods in predicting one's future companies and positions--i.e., on average, yielding 6.80% and 34.58% more accurate predictions, respectively. The codebase of CAPER is available at https://github.com/Bigdasgit/CAPER.

Jason Lucas, Adaku Uchendu, Michiharu Yamashita et al.

Recent ubiquity and disruptive impacts of large language models (LLMs) have raised concerns about their potential to be misused (.i.e, generating large-scale harmful and misleading content). To combat this emerging risk of LLMs, we propose a novel "Fighting Fire with Fire" (F3) strategy that harnesses modern LLMs' generative and emergent reasoning capabilities to counter human-written and LLM-generated disinformation. First, we leverage GPT-3.5-turbo to synthesize authentic and deceptive LLM-generated content through paraphrase-based and perturbation-based prefix-style prompts, respectively. Second, we apply zero-shot in-context semantic reasoning techniques with cloze-style prompts to discern genuine from deceptive posts and news articles. In our extensive experiments, we observe GPT-3.5-turbo's zero-shot superiority for both in-distribution and out-of-distribution datasets, where GPT-3.5-turbo consistently achieved accuracy at 68-72%, unlike the decline observed in previous customized and fine-tuned disinformation detectors. Our codebase and dataset are available at https://github.com/mickeymst/F3.

Saranya Venkatraman, Adaku Uchendu, Dongwon Lee

The Uniform Information Density (UID) principle posits that humans prefer to spread information evenly during language production. We examine if this UID principle can help capture differences between Large Language Models (LLMs)-generated and human-generated texts. We propose GPT-who, the first psycholinguistically-inspired domain-agnostic statistical detector. This detector employs UID-based features to model the unique statistical signature of each LLM and human author for accurate detection. We evaluate our method using 4 large-scale benchmark datasets and find that GPT-who outperforms state-of-the-art detectors (both statistical- & non-statistical) such as GLTR, GPTZero, DetectGPT, OpenAI detector, and ZeroGPT by over $20$% across domains. In addition to better performance, it is computationally inexpensive and utilizes an interpretable representation of text articles. We find that GPT-who can distinguish texts generated by very sophisticated LLMs, even when the overlying text is indiscernible. UID-based measures for all datasets and code are available at https://github.com/saranya-venkatraman/gpt-who.

Adaku Uchendu, Thai Le, Dongwon Lee

Recent advances in Large Language Models (LLMs) have enabled the generation of open-ended high-quality texts, that are non-trivial to distinguish from human-written texts. We refer to such LLM-generated texts as deepfake texts. There are currently over 72K text generation models in the huggingface model repo. As such, users with malicious intent can easily use these open-sourced LLMs to generate harmful texts and dis/misinformation at scale. To mitigate this problem, a computational method to determine if a given text is a deepfake text or not is desired--i.e., Turing Test (TT). In particular, in this work, we investigate the more general version of the problem, known as Authorship Attribution (AA), in a multi-class setting--i.e., not only determining if a given text is a deepfake text or not but also being able to pinpoint which LLM is the author. We propose TopFormer to improve existing AA solutions by capturing more linguistic patterns in deepfake texts by including a Topological Data Analysis (TDA) layer in the Transformer-based model. We show the benefits of having a TDA layer when dealing with imbalanced, and multi-style datasets, by extracting TDA features from the reshaped $pooled\_output$ of our backbone as input. This Transformer-based model captures contextual representations (i.e., semantic and syntactic linguistic features), while TDA captures the shape and structure of data (i.e., linguistic structures). Finally, TopFormer, outperforms all baselines in all 3 datasets, achieving up to 7\% increase in Macro F1 score. Our code and datasets are available at: https://github.com/AdaUchendu/topformer

Aashish Anantha Ramakrishnan, Sharon X. Huang, Dongwon Lee

Advancements in Text-to-Image synthesis over recent years have focused more on improving the quality of generated samples using datasets with descriptive prompts. However, real-world image-caption pairs present in domains such as news data do not use simple and directly descriptive captions. With captions containing information on both the image content and underlying contextual cues, they become abstractive in nature. In this paper, we launch ANNA, an Abstractive News captioNs dAtaset extracted from online news articles in a variety of different contexts. We explore the capabilities of current Text-to-Image synthesis models to generate news domain-specific images using abstractive captions by benchmarking them on ANNA, in both standard training and transfer learning settings. The generated images are judged on the basis of contextual relevance, visual quality, and perceptual similarity to ground-truth image-caption pairs. Through our experiments, we show that techniques such as transfer learning achieve limited success in understanding abstractive captions but still fail to consistently learn the relationships between content and context features. The Dataset is available at https://github.com/aashish2000/ANNA .

Adaku Uchendu, Thai Le, Dongwon Lee

Two interlocking research questions of growing interest and importance in privacy research are Authorship Attribution (AA) and Authorship Obfuscation (AO). Given an artifact, especially a text t in question, an AA solution aims to accurately attribute t to its true author out of many candidate authors while an AO solution aims to modify t to hide its true authorship. Traditionally, the notion of authorship and its accompanying privacy concern is only toward human authors. However, in recent years, due to the explosive advancements in Neural Text Generation (NTG) techniques in NLP, capable of synthesizing human-quality open-ended texts (so-called "neural texts"), one has to now consider authorships by humans, machines, or their combination. Due to the implications and potential threats of neural texts when used maliciously, it has become critical to understand the limitations of traditional AA/AO solutions and develop novel AA/AO solutions in dealing with neural texts. In this survey, therefore, we make a comprehensive review of recent literature on the attribution and obfuscation of neural text authorship from a Data Mining perspective, and share our view on their limitations and promising research directions.

Thai Le, Jooyoung Lee, Kevin Yen et al.

We proposes a novel algorithm, ANTHRO, that inductively extracts over 600K human-written text perturbations in the wild and leverages them for realistic adversarial attack. Unlike existing character-based attacks which often deductively hypothesize a set of manipulation strategies, our work is grounded on actual observations from real-world texts. We find that adversarial texts generated by ANTHRO achieve the best trade-off between (1) attack success rate, (2) semantic preservation of the original text, and (3) stealthiness--i.e. indistinguishable from human writings hence harder to be flagged as suspicious. Specifically, our attacks accomplished around 83% and 91% attack success rates on BERT and RoBERTa, respectively. Moreover, it outperformed the TextBugger baseline with an increase of 50% and 40% in terms of semantic preservation and stealthiness when evaluated by both layperson and professional human workers. ANTHRO can further enhance a BERT classifier's performance in understanding different variations of human-written toxic texts via adversarial training when compared to the Perspective API.

Nafis Irtiza Tripto, Saranya Venkatraman, Dominik Macko et al.

In the realm of text manipulation and linguistic transformation, the question of authorship has been a subject of fascination and philosophical inquiry. Much like the Ship of Theseus paradox, which ponders whether a ship remains the same when each of its original planks is replaced, our research delves into an intriguing question: Does a text retain its original authorship when it undergoes numerous paraphrasing iterations? Specifically, since Large Language Models (LLMs) have demonstrated remarkable proficiency in both the generation of original content and the modification of human-authored texts, a pivotal question emerges concerning the determination of authorship in instances where LLMs or similar paraphrasing tools are employed to rephrase the text--i.e., whether authorship should be attributed to the original human author or the AI-powered tool. Therefore, we embark on a philosophical voyage through the seas of language and authorship to unravel this intricate puzzle. Using a computational approach, we discover that the diminishing performance in text classification models, with each successive paraphrasing iteration, is closely associated with the extent of deviation from the original author's style, thus provoking a reconsideration of the current notion of authorship.

Xianren Zhang, Dongwon Lee, Suhang Wang

As deep vision models' popularity rapidly increases, there is a growing emphasis on explanations for model predictions. The inherently explainable attribution method aims to enhance the understanding of model behavior by identifying the important regions in images that significantly contribute to predictions. It is achieved by cooperatively training a selector (generating an attribution map to identify important features) and a predictor (making predictions using the identified features). Despite many advancements, existing methods suffer from the incompleteness problem, where discriminative features are masked out, and the interlocking problem, where the non-optimized selector initially selects noise, causing the predictor to fit on this noise and perpetuate the cycle. To address these problems, we introduce a new objective that discourages the presence of discriminative features in the masked-out regions thus enhancing the comprehensiveness of feature selection. A pre-trained detector is introduced to detect discriminative features in the masked-out region. If the selector selects noise instead of discriminative features, the detector can observe and break the interlocking situation by penalizing the selector. Extensive experiments show that our model makes accurate predictions with higher accuracy than the regular black-box model, and produces attribution maps with high feature coverage, localization ability, fidelity and robustness. Our code will be available at \href{https://github.com/Zood123/COMET}{https://github.com/Zood123/COMET}.

Yiran Ye, Thai Le, Dongwon Lee

Online texts with toxic content are a clear threat to the users on social media in particular and society in general. Although many platforms have adopted various measures (e.g., machine learning-based hate-speech detection systems) to diminish their effect, toxic content writers have also attempted to evade such measures by using cleverly modified toxic words, so-called human-written text perturbations. Therefore, to help build automatic detection tools to recognize those perturbations, prior methods have developed sophisticated techniques to generate diverse adversarial samples. However, we note that these ``algorithms"-generated perturbations do not necessarily capture all the traits of ``human"-written perturbations. Therefore, in this paper, we introduce a novel, high-quality dataset of human-written perturbations, named as NoisyHate, that was created from real-life perturbations that are both written and verified by human-in-the-loop. We show that perturbations in NoisyHate have different characteristics than prior algorithm-generated toxic datasets show, and thus can be in particular useful to help develop better toxic speech detection solutions. We thoroughly validate NoisyHate against state-of-the-art language models, such as BERT and RoBERTa, and black box APIs, such as Perspective API, on two tasks, such as perturbation normalization and understanding.

Mahjabin Nahar, Haeseung Seo, Eun-Ju Lee et al.

The widespread adoption and transformative effects of large language models (LLMs) have sparked concerns regarding their capacity to produce inaccurate and fictitious content, referred to as `hallucinations'. Given the potential risks associated with hallucinations, humans should be able to identify them. This research aims to understand the human perception of LLM hallucinations by systematically varying the degree of hallucination (genuine, minor hallucination, major hallucination) and examining its interaction with warning (i.e., a warning of potential inaccuracies: absent vs. present). Participants (N=419) from Prolific rated the perceived accuracy and engaged with content (e.g., like, dislike, share) in a Q/A format. Participants ranked content as truthful in the order of genuine, minor hallucination, and major hallucination, and user engagement behaviors mirrored this pattern. More importantly, we observed that warning improved the detection of hallucination without significantly affecting the perceived truthfulness of genuine content. We conclude by offering insights for future tools to aid human detection of hallucinations. All survey materials, demographic questions, and post-session questions are available at: https://github.com/MahjabinNahar/fakes-of-varying-shades-survey-materials

Eric Xing, Saranya Venkatraman, Thai Le et al.

Authorship Attribution (AA) and Authorship Obfuscation (AO) are two competing tasks of increasing importance in privacy research. Modern AA leverages an author's consistent writing style to match a text to its author using an AA classifier. AO is the corresponding adversarial task, aiming to modify a text in such a way that its semantics are preserved, yet an AA model cannot correctly infer its authorship. To address privacy concerns raised by state-of-the-art (SOTA) AA methods, new AO methods have been proposed but remain largely impractical to use due to their prohibitively slow training and obfuscation speed, often taking hours. To this challenge, we propose a practical AO method, ALISON, that (1) dramatically reduces training/obfuscation time, demonstrating more than 10x faster obfuscation than SOTA AO methods, (2) achieves better obfuscation success through attacking three transformer-based AA methods on two benchmark datasets, typically performing 15% better than competing methods, (3) does not require direct signals from a target AA classifier during obfuscation, and (4) utilizes unique stylometric features, allowing sound model interpretation for explainable obfuscation. We also demonstrate that ALISON can effectively prevent four SOTA AA methods from accurately determining the authorship of ChatGPT-generated texts, all while minimally changing the original text semantics. To ensure the reproducibility of our findings, our code and data are available at: https://github.com/EricX003/ALISON.

Aashish Anantha Ramakrishnan, Aadarsh Anantha Ramakrishnan, Dongwon Lee

Writing Assistants (e.g., Grammarly, Microsoft Copilot) traditionally generate diverse image captions by employing syntactic and semantic variations to describe image components. However, human-written captions prioritize conveying a central message alongside visual descriptions using pragmatic cues. To enhance caption diversity, it is essential to explore alternative ways of communicating these messages in conjunction with visual content. We propose RONA, a novel prompting strategy for Multi-modal Large Language Models (MLLM) that leverages Coherence Relations as a controllable axis for pragmatic variations. We demonstrate that RONA generates captions with better overall diversity and ground-truth alignment, compared to MLLM baselines across multiple domains. Our code is available at: https://github.com/aashish2000/RONA

Aashish Anantha Ramakrishnan, Aadarsh Anantha Ramakrishnan, Dongwon Lee

Interpreting figurative language such as sarcasm across multi-modal inputs presents unique challenges, often requiring task-specific fine-tuning and extensive reasoning steps. However, current Chain-of-Thought approaches do not efficiently leverage the same cognitive processes that enable humans to identify sarcasm. We present IRONIC, an in-context learning framework that leverages Multi-modal Coherence Relations to analyze referential, analogical and pragmatic image-text linkages. Our experiments show that IRONIC achieves state-of-the-art performance on zero-shot Multi-modal Sarcasm Detection across different baselines. This demonstrates the need for incorporating linguistic and cognitive insights into the design of multi-modal reasoning strategies. Our code is available at: https://github.com/aashish2000/IRONIC

Nafis Irtiza Tripto, Saranya Venkatraman, Mahjabin Nahar et al.

The rapid advancement of Large Language Models (LLMs) has revolutionized text generation but also raised concerns about potential misuse, making detecting LLM-generated text (AI text) increasingly essential. While prior work has focused on identifying AI text and effectively checkmating it, our study investigates a less-explored territory: portraying the nuanced distinctions between human and AI texts across text segments (introduction, body, and conclusion). Whether LLMs excel or falter in incorporating linguistic ingenuity across text segments, the results will critically inform their viability and boundaries as effective creative assistants to humans. Through an analogy with the structure of chess games, comprising opening, middle, and end games, we analyze segment-specific patterns to reveal where the most striking differences lie. Although AI texts closely resemble human writing in the body segment due to its length, deeper analysis shows a higher divergence in features dependent on the continuous flow of language, making it the most informative segment for detection. Additionally, human texts exhibit greater stylistic variation across segments, offering a new lens for distinguishing them from AI. Overall, our findings provide fresh insights into human-AI text differences and pave the way for more effective and interpretable detection strategies. Codes available at https://github.com/tripto03/chess_inspired_human_ai_text_distinction.

Jia Tracy Shen, Dongwon Lee

The issue of missing data poses a great challenge on boosting performance and application of deep learning models in the {\em Knowledge Tracing} (KT) problem. However, there has been the lack of understanding on the issue in the literature. %are not sufficient studies tackling this problem. In this work, to address this challenge, we adopt a subject-based training method to split and impute data by student IDs instead of row number splitting which we call non-subject based training. The benefit of subject-based training can retain the complete sequence for each student and hence achieve efficient training. Further, we leverage two existing deep generative frameworks, namely variational Autoencoders (VAE) and Longitudinal Variational Autoencoders (LVAE) frameworks and build LSTM kernels into them to form LSTM-VAE and LSTM LVAE (noted as VAE and LVAE for simplicity) models to generate quality data. In LVAE, a Gaussian Process (GP) model is trained to disentangle the correlation between the subject (i.e., student) descriptor information (e.g., age, gender) and the latent space. The paper finally compare the model performance between training the original data and training the data imputed with generated data from non-subject based model VAE-NS and subject-based training models (i.e., VAE and LVAE). We demonstrate that the generated data from LSTM-VAE and LSTM-LVAE can boost the original model performance by about 50%. Moreover, the original model just needs 10% more student data to surpass the original performance if the prediction model is small and 50\% more data if the prediction model is large with our proposed frameworks.

Jooyoung Lee, Toshini Agrawal, Adaku Uchendu et al.

Recent studies have raised concerns about the potential threats large language models (LLMs) pose to academic integrity and copyright protection. Yet, their investigation is predominantly focused on literal copies of original texts. Also, how LLMs can facilitate the detection of LLM-generated plagiarism remains largely unexplored. To address these gaps, we introduce \textbf{\sf PlagBench}, a dataset of 46.5K synthetic text pairs that represent three major types of plagiarism: verbatim copying, paraphrasing, and summarization. These samples are generated by three advanced LLMs. We rigorously validate the quality of PlagBench through a combination of fine-grained automatic evaluation and human annotation. We then utilize this dataset for two purposes: (1) to examine LLMs' ability to transform original content into accurate paraphrases and summaries, and (2) to evaluate the plagiarism detection performance of five modern LLMs alongside three specialized plagiarism checkers. Our results show that GPT-3.5 Turbo can produce high-quality paraphrases and summaries without significantly increasing text complexity compared to GPT-4 Turbo. However, in terms of detection, GPT-4 outperforms other LLMs and commercial detection tools by 20%, highlights the evolving capabilities of LLMs not only in content generation but also in plagiarism detection. Data and source code are available at https://github.com/Brit7777/plagbench.

Saranya Venkatraman, Nafis Irtiza Tripto, Dongwon Lee

The rise of unifying frameworks that enable seamless interoperability of Large Language Models (LLMs) has made LLM-LLM collaboration for open-ended tasks a possibility. Despite this, there have not been efforts to explore such collaborative writing. We take the next step beyond human-LLM collaboration to explore this multi-LLM scenario by generating the first exclusively LLM-generated collaborative stories dataset called CollabStory. We focus on single-author to multi-author (up to 5 LLMs) scenarios, where multiple LLMs co-author stories. We generate over 32k stories using open-source instruction-tuned LLMs. Further, we take inspiration from the PAN tasks that have set the standard for human-human multi-author writing tasks and analysis. We extend their authorship-related tasks for multi-LLM settings and present baselines for LLM-LLM collaboration. We find that current baselines are not able to handle this emerging scenario. Thus, CollabStory is a resource that could help propel an understanding as well as the development of new techniques to discern the use of multiple LLMs. This is crucial to study in the context of writing tasks since LLM-LLM collaboration could potentially overwhelm ongoing challenges related to plagiarism detection, credit assignment, maintaining academic integrity in educational settings, and addressing copyright infringement concerns. We make our dataset and code available at https://github.com/saranya-venkatraman/CollabStory.

Jia Tracy Shen, Michiharu Yamashita, Ethan Prihar et al.

Educational content labeled with proper knowledge components (KCs) are particularly useful to teachers or content organizers. However, manually labeling educational content is labor intensive and error-prone. To address this challenge, prior research proposed machine learning based solutions to auto-label educational content with limited success. In this work, we significantly improve prior research by (1) expanding the input types to include KC descriptions, instructional video titles, and problem descriptions (i.e., three types of prediction task), (2) doubling the granularity of the prediction from 198 to 385 KC labels (i.e., more practical setting but much harder multinomial classification problem), (3) improving the prediction accuracies by 0.5-2.3% using Task-adaptive Pre-trained BERT, outperforming six baselines, and (4) proposing a simple evaluation measure by which we can recover 56-73% of mispredicted KC labels. All codes and data sets in the experiments are available at:https://github.com/tbs17/TAPT-BERT

Limeng Cui, Dongwon Lee

As the COVID-19 virus quickly spreads around the world, unfortunately, misinformation related to COVID-19 also gets created and spreads like wild fire. Such misinformation has caused confusion among people, disruptions in society, and even deadly consequences in health problems. To be able to understand, detect, and mitigate such COVID-19 misinformation, therefore, has not only deep intellectual values but also huge societal impacts. To help researchers combat COVID-19 health misinformation, therefore, we present CoAID (Covid-19 heAlthcare mIsinformation Dataset), with diverse COVID-19 healthcare misinformation, including fake news on websites and social platforms, along with users' social engagement about such news. CoAID includes 4,251 news, 296,000 related user engagements, 926 social platform posts about COVID-19, and ground truth labels. The dataset is available at: https://github.com/cuilimeng/CoAID.

Dominik Macko, Robert Moro, Adaku Uchendu et al.

High-quality text generation capability of recent Large Language Models (LLMs) causes concerns about their misuse (e.g., in massive generation/spread of disinformation). Machine-generated text (MGT) detection is important to cope with such threats. However, it is susceptible to authorship obfuscation (AO) methods, such as paraphrasing, which can cause MGTs to evade detection. So far, this was evaluated only in monolingual settings. Thus, the susceptibility of recently proposed multilingual detectors is still unknown. We fill this gap by comprehensively benchmarking the performance of 10 well-known AO methods, attacking 37 MGT detection methods against MGTs in 11 languages (i.e., 10 $\times$ 37 $\times$ 11 = 4,070 combinations). We also evaluate the effect of data augmentation on adversarial robustness using obfuscated texts. The results indicate that all tested AO methods can cause evasion of automated detection in all tested languages, where homoglyph attacks are especially successful. However, some of the AO methods severely damaged the text, making it no longer readable or easily recognizable by humans (e.g., changed language, weird characters).

Delvin Ce Zhang, Dongwon Lee

Fact-checking the truthfulness of claims usually requires reasoning over multiple evidence sentences. Oftentimes, evidence sentences may not be always self-contained, and may require additional contexts and references from elsewhere to understand coreferential expressions, acronyms, and the scope of a reported finding. For example, evidence sentences from an academic paper may need contextual sentences in the paper and descriptions in its cited papers to determine the scope of a research discovery. However, most fact-checking models mainly focus on the reasoning within evidence sentences, and ignore the auxiliary contexts and references. To address this problem, we propose a novel method, Context- and Reference-augmented Reasoning and Prompting. For evidence reasoning, we construct a three-layer evidence graph with evidence, context, and reference layers. We design intra- and cross-layer reasoning to integrate three graph layers into a unified evidence embedding. For verdict prediction, we design evidence-conditioned prompt encoder, which produces unique prompt embeddings for each claim. These evidence-conditioned prompt embeddings and claims are unified for fact-checking. Experiments verify the strength of our model.

Aashish Anantha Ramakrishnan, Aadarsh Anantha Ramakrishnan, Dongwon Lee

Multimodal Large Language Models (MLLMs) are renowned for their superior instruction-following and reasoning capabilities across diverse problem domains. However, existing benchmarks primarily focus on assessing factual and logical correctness in downstream tasks, with limited emphasis on evaluating MLLMs' ability to interpret pragmatic cues and intermodal relationships. To address this gap, we assess the competency of MLLMs in performing Multimodal Discourse Analysis (MDA) using Coherence Relations. Our benchmark, CORDIAL, encompasses a broad spectrum of Coherence Relations across 3 different discourse domains at varying levels of granularity. Through our experiments on 10+ MLLMs employing different prompting strategies, we show that even top models like Gemini 1.5 Pro and GPT-4o fail to match the performance of simple classifier-based baselines. This study emphasizes the need to move beyond similarity-based metrics and adopt a discourse-driven framework for evaluating MLLMs, providing a more nuanced assessment of their capabilities. The benchmark and code are available at: https://aashish2000.github.io/CORDIAL/

Xianren Zhang, Xianfeng Tang, Hui Liu et al.

Recent studies show LLMs struggle with complex instructions involving multiple constraints (e.g., length, format, sentiment). Existing works address this issue by fine-tuning, which heavily relies on fine-tuning data quality and is computational expensive. An alternative is leveraging LLMs' self-correction to refine responses for better constraint adherence. However, this is limited by the feedback quality, as LLMs cannot generate reliable feedback or detect errors. Moreover, its effectiveness relies on few-shot examples illustrating response modifications. As constraints in complex instructions are diverse, manually crafting such examples for each constraint type can be labor-intensive and sub-optimal. To address these two challenges, we propose the Divide-Verify-Refine (DVR) framework with three steps: (1) Divide complex instructions into single constraints and prepare appropriate tools; (2) Verify responses using tools that provide rigorous check and textual guidance (e.g., Python toolkit for format checks or pre-trained classifiers for content analysis); (3) Refine: To maximize refinement effectiveness, we propose dynamic few-shot prompting, where a refinement repository collects successful refinements, and these examples are selectively retrieved for future refinements. Recognizing the lack of complexity in existing datasets, we create a new dataset of complex instructions. DVR doubles Llama3.1-8B's constraint adherence and triples Mistral-7B's performance.

Dominik Macko, Aashish Anantha Ramakrishnan, Jason Samuel Lucas et al.

Increased sophistication of large language models (LLMs) and the consequent quality of generated multilingual text raises concerns about potential disinformation misuse. While humans struggle to distinguish LLM-generated content from human-written texts, the scholarly debate about their impact remains divided. Some argue that heightened fears are overblown due to natural ecosystem limitations, while others contend that specific "longtail" contexts face overlooked risks. Our study bridges this debate by providing the first empirical evidence of LLM presence in the latest real-world disinformation datasets, documenting the increase of machine-generated content following ChatGPT's release, and revealing crucial patterns across languages, platforms, and time periods.

Wooyong Jung, Sola Kim, Dongwook Kim et al.

Homelessness in the United States has surged to levels unseen since the Great Depression. However, existing methods for monitoring it, such as point-in-time (PIT) counts, have limitations in terms of frequency, consistency, and spatial detail. This study proposes a new approach using publicly available, crowdsourced data, specifically 311 Service Calls and street-level imagery, to track and forecast homeless tent trends in San Francisco. Our predictive model captures fine-grained daily and neighborhood-level variations, uncovering patterns that traditional counts often overlook, such as rapid fluctuations during the COVID-19 pandemic and spatial shifts in tent locations over time. By providing more timely, localized, and cost-effective information, this approach serves as a valuable tool for guiding policy responses and evaluating interventions aimed at reducing unsheltered homelessness.

Xianren Zhang, Hui Liu, Delvin Ce Zhang et al.

Multimodal Large Language Models (MLLMs) trained on massive data may memorize sensitive personal information and photos, posing serious privacy risks. To mitigate this, MLLM unlearning methods are proposed, which fine-tune MLLMs to reduce the ``forget'' sensitive information. However, it remains unclear whether the knowledge has been truly forgotten or just hidden in the model. Therefore, we propose to study a novel problem of LLM unlearning attack, which aims to recover the unlearned knowledge of an unlearned LLM. To achieve the goal, we propose a novel framework Stealthy Unlearning Attack (SUA) framework that learns a universal noise pattern. When applied to input images, this noise can trigger the model to reveal unlearned content. While pixel-level perturbations may be visually subtle, they can be detected in the semantic embedding space, making such attacks vulnerable to potential defenses. To improve stealthiness, we introduce an embedding alignment loss that minimizes the difference between the perturbed and denoised image embeddings, ensuring the attack is semantically unnoticeable. Experimental results show that SUA can effectively recover unlearned information from MLLMs. Furthermore, the learned noise generalizes well: a single perturbation trained on a subset of samples can reveal forgotten content in unseen images. This indicates that knowledge reappearance is not an occasional failure, but a consistent behavior.

Hongchao Fang, Yixin Liu, Jiangshu Du et al.

AI-generated content is becoming increasingly prevalent in the real world, leading to serious ethical and societal concerns. For instance, adversaries might exploit large multimodal models (LMMs) to create images that violate ethical or legal standards, while paper reviewers may misuse large language models (LLMs) to generate reviews without genuine intellectual effort. While prior work has explored detecting AI-generated images and texts, and occasionally tracing their source models, there is a lack of a systematic and fine-grained comparative study. Important dimensions--such as AI-generated images vs. text, fully vs. partially AI-generated images, and general vs. malicious use cases--remain underexplored. Furthermore, whether AI systems like GPT-4o can explain why certain forged content is attributed to specific generative models is still an open question, with no existing benchmark addressing this. To fill this gap, we introduce AI-FAKER, a comprehensive multimodal dataset with over 280,000 samples spanning multiple LLMs and LMMs, covering both general and malicious use cases for AI-generated images and texts. Our experiments reveal two key findings: (i) AI authorship detection depends not only on the generated output but also on the model's original training intent; and (ii) GPT-4o provides highly consistent but less specific explanations when analyzing content produced by OpenAI's own models, such as DALL-E and GPT-4o itself.

Harsh Nishant Lalai, Aashish Anantha Ramakrishnan, Raj Sanjay Shah et al.

With the rapid growth of Large Language Models (LLMs), safeguarding textual content against unauthorized use is crucial. Watermarking offers a vital solution, protecting both - LLM-generated and plain text sources. This paper presents a unified overview of different perspectives behind designing watermarking techniques through a comprehensive survey of the research literature. Our work has two key advantages: (1) We analyze research based on the specific intentions behind different watermarking techniques, evaluation datasets used, and watermarking addition and removal methods to construct a cohesive taxonomy. (2) We highlight the gaps and open challenges in text watermarking to promote research protecting text authorship. This extensive coverage and detailed analysis sets our work apart, outlining the evolving landscape of text watermarking in Language Models.

Aashish Anantha Ramakrishnan, Sharon X. Huang, Dongwon Lee

Text-to-Image (T2I) Synthesis has made tremendous strides in enhancing synthesized image quality, but current datasets evaluate model performance only on descriptive, instruction-based prompts. Real-world news image captions take a more pragmatic approach, providing high-level situational and Named-Entity (NE) information and limited physical object descriptions, making them abstractive. To evaluate the ability of T2I models to capture intended subjects from news captions, we introduce the Abstractive News Captions with High-level cOntext Representation (ANCHOR) dataset, containing 70K+ samples sourced from 5 different news media organizations. With Large Language Models (LLM) achieving success in language and commonsense reasoning tasks, we explore the ability of different LLMs to identify and understand key subjects from abstractive captions. Our proposed method Subject-Aware Finetuning (SAFE), selects and enhances the representation of key subjects in synthesized images by leveraging LLM-generated subject weights. It also adapts to the domain distribution of news images and captions through custom Domain Fine-tuning, outperforming current T2I baselines on ANCHOR. By launching the ANCHOR dataset, we hope to motivate research in furthering the Natural Language Understanding (NLU) capabilities of T2I models.

Dominik Macko, Robert Moro, Adaku Uchendu et al.

There is a lack of research into capabilities of recent LLMs to generate convincing text in languages other than English and into performance of detectors of machine-generated text in multilingual settings. This is also reflected in the available benchmarks which lack authentic texts in languages other than English and predominantly cover older generators. To fill this gap, we introduce MULTITuDE, a novel benchmarking dataset for multilingual machine-generated text detection comprising of 74,081 authentic and machine-generated texts in 11 languages (ar, ca, cs, de, en, es, nl, pt, ru, uk, and zh) generated by 8 multilingual LLMs. Using this benchmark, we compare the performance of zero-shot (statistical and black-box) and fine-tuned detectors. Considering the multilinguality, we evaluate 1) how these detectors generalize to unseen languages (linguistically similar as well as dissimilar) and unseen LLMs and 2) whether the detectors improve their performance when trained on multiple languages.

Michiharu Yamashita, Jia Tracy Shen, Thanh Tran et al.

In online job marketplaces, it is important to establish a well-defined job title taxonomy for various downstream tasks (e.g., job recommendation, users' career analysis, and turnover prediction). Job Title Normalization (JTN) is such a cleaning step to classify user-created non-standard job titles into normalized ones. However, solving the JTN problem is non-trivial with challenges: (1) semantic similarity of different job titles, (2) non-normalized user-created job titles, and (3) large-scale and long-tailed job titles in real-world applications. To this end, we propose a novel solution, named JAMES, that constructs three unique embeddings (i.e., graph, contextual, and syntactic) of a target job title to effectively capture its various traits. We further propose a multi-aspect co-attention mechanism to attentively combine these embeddings, and employ neural logical reasoning representations to collaboratively estimate similarities between messy job titles and normalized job titles in a reasoning space. To evaluate JAMES, we conduct comprehensive experiments against ten competing models on a large-scale real-world dataset with over 350,000 job titles. Our experimental results show that JAMES significantly outperforms the best baseline by 10.06% in Precision@10 and by 17.52% in NDCG@10, respectively.

Thai Le, Long Tran-Thanh, Dongwon Lee

Socialbots are software-driven user accounts on social platforms, acting autonomously (mimicking human behavior), with the aims to influence the opinions of other users or spread targeted misinformation for particular goals. As socialbots undermine the ecosystem of social platforms, they are often considered harmful. As such, there have been several computational efforts to auto-detect the socialbots. However, to our best knowledge, the adversarial nature of these socialbots has not yet been studied. This begs a question "can adversaries, controlling socialbots, exploit AI techniques to their advantage?" To this question, we successfully demonstrate that indeed it is possible for adversaries to exploit computational learning mechanism such as reinforcement learning (RL) to maximize the influence of socialbots while avoiding being detected. We first formulate the adversarial socialbot learning as a cooperative game between two functional hierarchical RL agents. While one agent curates a sequence of activities that can avoid the detection, the other agent aims to maximize network influence by selectively connecting with right users. Our proposed policy networks train with a vast amount of synthetic graphs and generalize better than baselines on unseen real-life graphs both in terms of maximizing network influence (up to +18%) and sustainable stealthiness (up to +40% undetectability) under a strong bot detector (with 90% detection accuracy). During inference, the complexity of our approach scales linearly, independent of a network's structure and the virality of news. This makes our approach a practical adversarial attack when deployed in a real-life setting.

Adaku Uchendu, Zeyu Ma, Thai Le et al.

Recent progress in generative language models has enabled machines to generate astonishingly realistic texts. While there are many legitimate applications of such models, there is also a rising need to distinguish machine-generated texts from human-written ones (e.g., fake news detection). However, to our best knowledge, there is currently no benchmark environment with datasets and tasks to systematically study the so-called "Turing Test" problem for neural text generation methods. In this work, we present the TuringBench benchmark environment, which is comprised of (1) a dataset with 200K human- or machine-generated samples across 20 labels {Human, GPT-1, GPT-2_small, GPT-2_medium, GPT-2_large, GPT-2_xl, GPT-2_PyTorch, GPT-3, GROVER_base, GROVER_large, GROVER_mega, CTRL, XLM, XLNET_base, XLNET_large, FAIR_wmt19, FAIR_wmt20, TRANSFORMER_XL, PPLM_distil, PPLM_gpt2}, (2) two benchmark tasks -- i.e., Turing Test (TT) and Authorship Attribution (AA), and (3) a website with leaderboards. Our preliminary experimental results using TuringBench show that FAIR_wmt20 and GPT-3 are the current winners, among all language models tested, in generating the most human-like indistinguishable texts with the lowest F1 score by five state-of-the-art TT detection models. The TuringBench is available at: https://turingbench.ist.psu.edu/

Minjin Choi, Sunkyung Lee, Eunseong Choi et al.

Automated metaphor detection is a challenging task to identify metaphorical expressions of words in a sentence. To tackle this problem, we adopt pre-trained contextualized models, e.g., BERT and RoBERTa. To this end, we propose a novel metaphor detection model, namely metaphor-aware late interaction over BERT (MelBERT). Our model not only leverages contextualized word representation but also benefits from linguistic metaphor identification theories to distinguish between the contextual and literal meaning of words. Our empirical results demonstrate that MelBERT outperforms several strong baselines on four benchmark datasets, i.e., VUA-18, VUA-20, MOH-X, and TroFi.

Thai Le, Noseong Park, Dongwon Lee

The Universal Trigger (UniTrigger) is a recently-proposed powerful adversarial textual attack method. Utilizing a learning-based mechanism, UniTrigger generates a fixed phrase that, when added to any benign inputs, can drop the prediction accuracy of a textual neural network (NN) model to near zero on a target class. To defend against this attack that can cause significant harm, in this paper, we borrow the "honeypot" concept from the cybersecurity community and propose DARCY, a honeypot-based defense framework against UniTrigger. DARCY greedily searches and injects multiple trapdoors into an NN model to "bait and catch" potential attacks. Through comprehensive experiments across four public datasets, we show that DARCY detects UniTrigger's adversarial attacks with up to 99% TPR and less than 2% FPR in most cases, while maintaining the prediction accuracy (in F1) for clean inputs within a 1% margin. We also demonstrate that DARCY with multiple trapdoors is also robust to a diverse set of attack scenarios with attackers' varying levels of knowledge and skills. Source code will be released upon the acceptance of this paper.

Thai Le, Noseong Park, Dongwon Lee

Even though several methods have proposed to defend textual neural network (NN) models against black-box adversarial attacks, they often defend against a specific text perturbation strategy and/or require re-training the models from scratch. This leads to a lack of generalization in practice and redundant computation. In particular, the state-of-the-art transformer models (e.g., BERT, RoBERTa) require great time and computation resources. By borrowing an idea from software engineering, in order to address these limitations, we propose a novel algorithm, SHIELD, which modifies and re-trains only the last layer of a textual NN, and thus it "patches" and "transforms" the NN into a stochastic weighted ensemble of multi-expert prediction heads. Considering that most of current black-box attacks rely on iterative search mechanisms to optimize their adversarial perturbations, SHIELD confuses the attackers by automatically utilizing different weighted ensembles of predictors depending on the input. In other words, SHIELD breaks a fundamental assumption of the attack, which is a victim NN model remains constant during an attack. By conducting comprehensive experiments, we demonstrate that all of CNN, RNN, BERT, and RoBERTa-based textual NNs, once patched by SHIELD, exhibit a relative enhancement of 15%--70% in accuracy on average against 14 different black-box attacks, outperforming 6 defensive baselines across 3 public datasets. All codes are to be released.

Wen Huang, Kevin Labille, Xintao Wu et al.

Personalized recommendation based on multi-arm bandit (MAB) algorithms has shown to lead to high utility and efficiency as it can dynamically adapt the recommendation strategy based on feedback. However, unfairness could incur in personalized recommendation. In this paper, we study how to achieve user-side fairness in personalized recommendation. We formulate our fair personalized recommendation as a modified contextual bandit and focus on achieving fairness on the individual whom is being recommended an item as opposed to achieving fairness on the items that are being recommended. We introduce and define a metric that captures the fairness in terms of rewards received for both the privileged and protected groups. We develop a fair contextual bandit algorithm, Fair-LinUCB, that improves upon the traditional LinUCB algorithm to achieve group-level fairness of users. Our algorithm detects and monitors unfairness while it learns to recommend personalized videos to students to achieve high efficiency. We provide a theoretical regret analysis and show that our algorithm has a slightly higher regret bound than LinUCB. We conduct numerous experimental evaluations to compare the performances of our fair contextual bandit to that of LinUCB and show that our approach achieves group-level fairness while maintaining a high utility.

Thai Le, Suhang Wang, Dongwon Lee

In recent years, the proliferation of so-called "fake news" has caused much disruptions in society and weakened the news ecosystem. Therefore, to mitigate such problems, researchers have developed state-of-the-art models to auto-detect fake news on social media using sophisticated data science and machine learning techniques. In this work, then, we ask "what if adversaries attempt to attack such detection models?" and investigate related issues by (i) proposing a novel threat model against fake news detectors, in which adversaries can post malicious comments toward news articles to mislead fake news detectors, and (ii) developing MALCOM, an end-to-end adversarial comment generation framework to achieve such an attack. Through a comprehensive evaluation, we demonstrate that about 94% and 93.5% of the time on average MALCOM can successfully mislead five of the latest neural detection models to always output targeted real and fake news labels. Furthermore, MALCOM can also fool black box fake news detectors to always output real news labels 90% of the time on average. We also compare our attack model with four baselines across two real-world datasets, not only on attack performance but also on generated quality, coherency, transferability, and robustness.

Kai Shu, Suhang Wang, Dongwon Lee et al.

In recent years, disinformation including fake news, has became a global phenomenon due to its explosive growth, particularly on social media. The wide spread of disinformation and fake news can cause detrimental societal effects. Despite the recent progress in detecting disinformation and fake news, it is still non-trivial due to its complexity, diversity, multi-modality, and costs of fact-checking or annotation. The goal of this chapter is to pave the way for appreciating the challenges and advancements via: (1) introducing the types of information disorder on social media and examine their differences and connections; (2) describing important and emerging tasks to combat disinformation for characterization, detection and attribution; and (3) discussing a weak supervision approach to detect disinformation with limited labeled data. We then provide an overview of the chapters in this book that represent the recent advancements in three related parts: (1) user engagements in the dissemination of information disorder; (2) techniques on detecting and mitigating disinformation; and (3) trending issues such as ethics, blockchain, clickbaits, etc. We hope this book to be a convenient entry point for researchers, practitioners, and students to understand the problems and challenges, learn state-of-the-art solutions for their specific needs, and quickly identify new research problems in their domains.

Thai Le, Suhang Wang, Dongwon Lee

Despite the recent development in the topic of explainable AI/ML for image and text data, the majority of current solutions are not suitable to explain the prediction of neural network models when the datasets are tabular and their features are in high-dimensional vectorized formats. To mitigate this limitation, therefore, we borrow two notable ideas (i.e., "explanation by intervention" from causality and "explanation are contrastive" from philosophy) and propose a novel solution, named as GRACE, that better explains neural network models' predictions for tabular datasets. In particular, given a model's prediction as label X, GRACE intervenes and generates a minimally-modified contrastive sample to be classified as Y, with an intuitive textual explanation, answering the question of "Why X rather than Y?" We carry out comprehensive experiments using eleven public datasets of different scales and domains (e.g., # of features ranges from 5 to 216) and compare GRACE with competing baselines on different measures: fidelity, conciseness, info-gain, and influence. The user-studies show that our generated explanation is not only more intuitive and easy-to-understand but also facilitates end-users to make as much as 60% more accurate post-explanation decisions than that of Lime.

Jason, Zhang, Junming Yin et al.

In recent years, \emph{search story}, a combined display with other organic channels, has become a major source of user traffic on platforms such as e-commerce search platforms, news feed platforms and web and image search platforms. The recommended search story guides a user to identify her own preference and personal intent, which subsequently influences the user's real-time and long-term search behavior. %With such an increased importance of search stories, As search stories become increasingly important, in this work, we study the problem of personalized search story recommendation within a search engine, which aims to suggest a search story relevant to both a search keyword and an individual user's interest. To address the challenge of modeling both immediate and future values of recommended search stories (i.e., cross-channel effect), for which conventional supervised learning framework is not applicable, we resort to a Markov decision process and propose a deep reinforcement learning architecture trained by both imitation learning and reinforcement learning. We empirically demonstrate the effectiveness of our proposed approach through extensive experiments on real-world data sets from JD.com.

Changtao Zhong, Hau-wen Chan, Dmytro Karamshuk et al.

This paper investigates when users create profiles in different social networks, whether they are redundant expressions of the same persona, or they are adapted to each platform. Using the personal webpages of 116,998 users on About.me, we identify and extract matched user profiles on several major social networks including Facebook, Twitter, LinkedIn, and Instagram. We find evidence for distinct site-specific norms, such as differences in the language used in the text of the profile self-description, and the kind of picture used as profile image. By learning a model that robustly identifies the platform given a user's profile image (0.657--0.829 AUC) or self-description (0.608--0.847 AUC), we confirm that users do adapt their behaviour to individual platforms in an identifiable and learnable manner. However, different genders and age groups adapt their behaviour differently from each other, and these differences are, in general, consistent across different platforms. We show that differences in social profile construction correspond to differences in how formal or informal the platform is.