Yusuf Mücahit Çetinkaya
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Yusuf Mücahit Çetinkaya has a total of 18 co-authors. Here are the top co-authors by number of shared publications:
- Hasan Davulcu
Arizona State University
5 shared publications - Emre Külah
Computer Engineering Department, Middle East Technical University, Ankara, Trkiye
4 shared publications - Ismail Hakki Toroslu
Department of Computer Engineering, Middle East Technical University, Ankara, Trkiye
4 shared publications - Guillermo Suarez-Tangil
IMDEA Networks Institute, Legans, Spain
2 shared publications - Jose Such
INGENIO (CSIC-UPV), Valencia, Spain
2 shared publications Michael A. Cowan
New Orleans, LA, USA
2 shared publications
Latest Publications
- 10 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet Computing
- Vol. PP (10897801) , 1-10
Evaluating Image Forgery Detection and Propaganda in Manipulated Images of the Russian Trolls Campaign
Generative AI tools have increased the scale and accessibility of deceptive image creation, but the robustness of existing forgery detection methods under emerging editing tools and real-world social media conditions remains uncertain. This study evaluates state-of-the-art image forgery detectors on recent benchmarks and applies the strongest model to a historical influence-campaign case study: Russian troll activity during the 2016 U.S. presidential election. We find that TruFor is the most consistent detector across legacy and recent datasets, but that its performance degrades on edits produced by advanced tools such as Nano Banana. We then deploy a multimodal pipeline combining forgery localization, Optical Character Recognition (OCR), face recognition, and visual-language-model-based annotation to characterize manipulated images in the campaign. Our analysis suggests that Russian troll accounts relied primarily on low-effort meme-style edits and ridicule-oriented reputation attacks, disproportionately targeting prominent U.S. political figures including Hillary Clinton, Bill Clinton, and Barack Obama.
- 10 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet Computing
- Vol. PP (10897801) , 1-10
Evaluating Image Forgery Detection and Propaganda in Manipulated Images of the Russian Trolls Campaign
Generative AI tools have increased the scale and accessibility of deceptive image creation, but the robustness of existing forgery detection methods under emerging editing tools and real-world social media conditions remains uncertain. This study evaluates state-of-the-art image forgery detectors on recent benchmarks and applies the strongest model to a historical influence-campaign case study: Russian troll activity during the 2016 U.S. presidential election. We find that TruFor is the most consistent detector across legacy and recent datasets, but that its performance degrades on edits produced by advanced tools such as Nano Banana. We then deploy a multimodal pipeline combining forgery localization, Optical Character Recognition (OCR), face recognition, and visual-language-model-based annotation to characterize manipulated images in the campaign. Our analysis suggests that Russian troll accounts relied primarily on low-effort meme-style edits and ridicule-oriented reputation attacks, disproportionately targeting prominent U.S. political figures including Hillary Clinton, Bill Clinton, and Barack Obama.
- 1 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. PP, 1-12
Beyond the Black Box: Programmable AI and Explainable Text Analysis for Trustworthy Social Intelligence
Life demands safe, reliable, and explainable artificial intelligence (AI) grounded in scientific reasoning and guided by human oversight. We present a neuro-symbolic programmable AI algorithm that transforms neural outputs into interpretable logic-patterned narratives via symbolic pattern lifting and lowering. It is designed for social scientists and enables reproducible Human + AI text analysis across sources such as social media, interviews, and news. It is demonstrated on 1.3 million tweets about global energy trends, and the system ensures AI that not only says what it does but does what it says, replacing black-box opacity with transparent, accountable reasoning.
- 1 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Computational Social Systems
- Vol. PP, 1-12
Beyond the Black Box: Programmable AI and Explainable Text Analysis for Trustworthy Social Intelligence
Life demands safe, reliable, and explainable artificial intelligence (AI) grounded in scientific reasoning and guided by human oversight. We present a neuro-symbolic programmable AI algorithm that transforms neural outputs into interpretable logic-patterned narratives via symbolic pattern lifting and lowering. It is designed for social scientists and enables reproducible Human + AI text analysis across sources such as social media, interviews, and news. It is demonstrated on 1.3 million tweets about global energy trends, and the system ensures AI that not only says what it does but does what it says, replacing black-box opacity with transparent, accountable reasoning.
- Open Access
- 14 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 74640-74659
SALDIRAY: Scalable and Adaptive Language Diagnostics for Remediation of Anomalies and Typos
Natural language processing (NLP) systems working with real-world user data often encounter noisy, informal text, including typos, spelling variations, slang, and homophone substitutions. This noise can cause a severe drop in the performance of downstream models. This paper presents SALDIRAY (Scalable and Adaptive Language Diagnostics for Remediation of Anomalies and Typos), an integrated, domain-independent framework for text normalization that aims to boost the resilience of NLP systems across various model architectures and application types. It converts short, poorly formed text into standardized language through a synthetic noise modeling and standardization approach, instantiated in practice as a noise generation and normalization pipeline. This approach is enriched by incorporating curated slang and homophone lists to ensure linguistic realism. The process creates large amounts of paired noisy–clean data, which are used to fine-tune robust normalization models. Our detailed experiments consistently show that SALDIRAYimproves the accuracy and stability of systems across a wide range of tasks, including traditional machine learning–based classification, sentiment analysis, and instruction-tuned large language model tasks. The framework requires no task-specific supervision, operates as a modular preprocessing layer, and can be integrated into existing NLP workflows. The experiments confirm that the proposed framework offers a scalable and adaptive method for improving robustness to noisy text, particularly under controlled corruption settings, while showing promising transfer to real-world noisy inputs.
- 14 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 74640-74659
Open AccessSALDIRAY: Scalable and Adaptive Language Diagnostics for Remediation of Anomalies and Typos
Natural language processing (NLP) systems working with real-world user data often encounter noisy, informal text, including typos, spelling variations, slang, and homophone substitutions. This noise can cause a severe drop in the performance of downstream models. This paper presents SALDIRAY (Scalable and Adaptive Language Diagnostics for Remediation of Anomalies and Typos), an integrated, domain-independent framework for text normalization that aims to boost the resilience of NLP systems across various model architectures and application types. It converts short, poorly formed text into standardized language through a synthetic noise modeling and standardization approach, instantiated in practice as a noise generation and normalization pipeline. This approach is enriched by incorporating curated slang and homophone lists to ensure linguistic realism. The process creates large amounts of paired noisy–clean data, which are used to fine-tune robust normalization models. Our detailed experiments consistently show that SALDIRAYimproves the accuracy and stability of systems across a wide range of tasks, including traditional machine learning–based classification, sentiment analysis, and instruction-tuned large language model tasks. The framework requires no task-specific supervision, operates as a modular preprocessing layer, and can be integrated into existing NLP workflows. The experiments confirm that the proposed framework offers a scalable and adaptive method for improving robustness to noisy text, particularly under controlled corruption settings, while showing promising transfer to real-world noisy inputs.
- Open Access
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 4623-4627
State & Geopolitical Censorship on Twitter (X): Detection & Impact Analysis of Withheld Content
- 10 November 2025
- Published by Association for Computing Machinery (ACM)
- p. 4623-4627
Open AccessState & Geopolitical Censorship on Twitter (X): Detection & Impact Analysis of Withheld Content
- Open Access
- 7 June 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the International AAAI Conference on Web and Social Media
- Vol. 19, 324-340
Cross-Partisan Interactions on Twitter
Many social media studies argue that social media creates echo chambers where some users only interact with peers of the same political orientation. However, recent studies suggest that a substantial amount of Cross-Partisan Interactions (CPIs) do exist --- even within echo chambers, but they may be toxic. There is no consensus about how such interactions occur and when they lead to healthy or toxic dialogue. In this paper, we study a comprehensive Twitter dataset that consists of 3 million tweets from 2020 related to the U.S. context to understand the dynamics behind CPIs. We investigate factors that are more associated with such interactions, including how users engage in CPIs, which topics are more contentious, and what are the stances associated with healthy interactions. We find that CPIs are significantly influenced by the nature of the topics being discussed, with politically charged events acting as strong catalysts. The political discourse and pre-established political views sway how users participate in CPIs, but the direction in which users go is nuanced. While Democrats engage in cross-partisan interactions slightly more frequently, these interactions often involve more negative and nonconstructive stances compared to their intra-party interactions. In contrast, Republicans tend to maintain a more consistent tone across interactions. Although users are more likely to engage in CPIs with popular accounts in general, this is less common among Republicans who often engage in CPIs with accounts with a low number of followers for personal matters. Our study has implications beyond Twitter as identifying topics with low toxicity and high CPI can help highlight potential opportunities for reducing polarization while topics with high toxicity and low CPI may action targeted interventions when moderating harm.
- 7 June 2025
- Published by Association for the Advancement of Artificial Intelligence (AAAI) in Proceedings of the International AAAI Conference on Web and Social Media
- Vol. 19, 324-340
Open AccessCross-Partisan Interactions on Twitter
Many social media studies argue that social media creates echo chambers where some users only interact with peers of the same political orientation. However, recent studies suggest that a substantial amount of Cross-Partisan Interactions (CPIs) do exist --- even within echo chambers, but they may be toxic. There is no consensus about how such interactions occur and when they lead to healthy or toxic dialogue. In this paper, we study a comprehensive Twitter dataset that consists of 3 million tweets from 2020 related to the U.S. context to understand the dynamics behind CPIs. We investigate factors that are more associated with such interactions, including how users engage in CPIs, which topics are more contentious, and what are the stances associated with healthy interactions. We find that CPIs are significantly influenced by the nature of the topics being discussed, with politically charged events acting as strong catalysts. The political discourse and pre-established political views sway how users participate in CPIs, but the direction in which users go is nuanced. While Democrats engage in cross-partisan interactions slightly more frequently, these interactions often involve more negative and nonconstructive stances compared to their intra-party interactions. In contrast, Republicans tend to maintain a more consistent tone across interactions. Although users are more likely to engage in CPIs with popular accounts in general, this is less common among Republicans who often engage in CPIs with accounts with a low number of followers for personal matters. Our study has implications beyond Twitter as identifying topics with low toxicity and high CPI can help highlight potential opportunities for reducing polarization while topics with high toxicity and low CPI may action targeted interventions when moderating harm.
- Open Access
Cross-Partisan Interactions on Twitter
Many social media studies argue that social media creates echo chambers where some users only interact with peers of the same political orientation. However, recent studies suggest that a substantial amount of Cross-Partisan Interactions (CPIs) do exist - even within echo chambers, but they may be toxic. There is no consensus about how such interactions occur and when they lead to healthy or toxic dialogue. In this paper, we study a comprehensive Twitter dataset that consists of 3 million tweets from 2020 related to the U.S. context to understand the dynamics behind CPIs. We investigate factors that are more associated with such interactions, including how users engage in CPIs, which topics are more contentious, and what are the stances associated with healthy interactions. We find that CPIs are significantly influenced by the nature of the topics being discussed, with politically charged events acting as strong catalysts. The political discourse and pre-established political views sway how users participate in CPIs, but the direction in which users go is nuanced. While Democrats engage in cross-partisan interactions slightly more frequently, these interactions often involve more negative and nonconstructive stances compared to their intra-party interactions. In contrast, Republicans tend to maintain a more consistent tone across interactions. Although users are more likely to engage in CPIs with popular accounts in general, this is less common among Republicans who often engage in CPIs with accounts with a low number of followers for personal matters. Our study has implications beyond Twitter as identifying topics with low toxicity and high CPI can help highlight potential opportunities for reducing polarization while topics with high toxicity and low CPI may action targeted interventions when moderating harm.
Open AccessCross-Partisan Interactions on Twitter
Many social media studies argue that social media creates echo chambers where some users only interact with peers of the same political orientation. However, recent studies suggest that a substantial amount of Cross-Partisan Interactions (CPIs) do exist - even within echo chambers, but they may be toxic. There is no consensus about how such interactions occur and when they lead to healthy or toxic dialogue. In this paper, we study a comprehensive Twitter dataset that consists of 3 million tweets from 2020 related to the U.S. context to understand the dynamics behind CPIs. We investigate factors that are more associated with such interactions, including how users engage in CPIs, which topics are more contentious, and what are the stances associated with healthy interactions. We find that CPIs are significantly influenced by the nature of the topics being discussed, with politically charged events acting as strong catalysts. The political discourse and pre-established political views sway how users participate in CPIs, but the direction in which users go is nuanced. While Democrats engage in cross-partisan interactions slightly more frequently, these interactions often involve more negative and nonconstructive stances compared to their intra-party interactions. In contrast, Republicans tend to maintain a more consistent tone across interactions. Although users are more likely to engage in CPIs with popular accounts in general, this is less common among Republicans who often engage in CPIs with accounts with a low number of followers for personal matters. Our study has implications beyond Twitter as identifying topics with low toxicity and high CPI can help highlight potential opportunities for reducing polarization while topics with high toxicity and low CPI may action targeted interventions when moderating harm.
- 25 January 2025
- Published by Springer Nature
- p. 214-225
Masking the Bias: From Echo Chambers to Large Scale Aspect-Based Sentiment Analysis
- 25 January 2025
- Published by Springer Nature
- p. 214-225
Masking the Bias: From Echo Chambers to Large Scale Aspect-Based Sentiment Analysis
- 29 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet Computing
- Vol. 28 (5) , 20-27
Toward a Programmable Humanizing Artificial Intelligence Through Scalable Stance-Directed Architecture
The rise of harmful online content underscores the urgent need for AI systems to effectively detect, filter those, and foster safer and healthier communication. This article introduces a novel approach to mitigate toxic content generation propensities of Large Language Models (LLMs) by fine-tuning them with a programmable stance-directed focus on core human values and common good. We propose a streamlined keyword coding and processing pipeline to generate weakly labeled data to train AI models that can avoid toxicity and champion civil discourse. We also developed a toxicity classifier and an Aspect-based Sentiment Analysis (ABSA) model to assess and control the effectiveness of a humanizing AI model. We evaluate the proposed pipeline using a contentious real-world Twitter dataset on U.S. race relations. Our approach successfully curbs the toxic content generation propensity of an unrestricted LLM by a significant 85%.
- 29 August 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet Computing
- Vol. 28 (5) , 20-27
Toward a Programmable Humanizing Artificial Intelligence Through Scalable Stance-Directed Architecture
The rise of harmful online content underscores the urgent need for AI systems to effectively detect, filter those, and foster safer and healthier communication. This article introduces a novel approach to mitigate toxic content generation propensities of Large Language Models (LLMs) by fine-tuning them with a programmable stance-directed focus on core human values and common good. We propose a streamlined keyword coding and processing pipeline to generate weakly labeled data to train AI models that can avoid toxicity and champion civil discourse. We also developed a toxicity classifier and an Aspect-based Sentiment Analysis (ABSA) model to assess and control the effectiveness of a humanizing AI model. We evaluate the proposed pipeline using a contentious real-world Twitter dataset on U.S. race relations. Our approach successfully curbs the toxic content generation propensity of an unrestricted LLM by a significant 85%.
- 4 April 2024
- Published by Springer Nature in Social Network Analysis and Mining
- Vol. 14 (1) , 1-15
Targeted marketing on social media: utilizing text analysis to create personalized landing pages
- 4 April 2024
- Published by Springer Nature in Social Network Analysis and Mining
- Vol. 14 (1) , 1-15
Targeted marketing on social media: utilizing text analysis to create personalized landing pages
- Open Access
- 27 March 2023
- Published by Springer Nature in Research Square
Targeted Marketing on Social Media: Utilizing Text Analysis to Create Personalized Landing Pages
The proliferation of social media has rendered it a critical arena for online marketing strategies. To optimize conversion rates, the landing pages must effectively respond to a visitor segment's pain points that they need solutions for. A one-size-fits-all approach is inadequate since even if the product meets the needs of all consumers, their priorities may diverge. In this study, we propose a pipeline for creating personalized landing pages that dynamically cater to visiting customers' specific concerns. As a use case, a pipeline will be utilized to create a personalized pharmacy discount card landing page, serving for the particular needs of chronic diabetics users seeking to purchase needed medications at a reduced cost. The proposed pipeline incorporates additional stages to augment the traditional online marketing funnel including acquisition of salient tweets, filtration of irrelevant ones, extraction of themes from relevant tweets, and generation of coherent paragraphs. To collect relevant tweets and reduce bias, Facebook groups and pages relevant to individuals with diabetes were leveraged. Pre-trained models such as BERT and RoBERTa were used to cluster the tweets based on their similarities. GuidedLDA exhibited superior performance in identifying customer priorities. Human evaluations revealed that personalized landing pages were more effective in getting the attention, building attraction by addressing their concerns and engaging the audiences. The proposed methodology offers a practical architecture for developing customized landing pages considering visiting customers' profiles and needs.
- 27 March 2023
- Published by Springer Nature in Research Square
Open AccessTargeted Marketing on Social Media: Utilizing Text Analysis to Create Personalized Landing Pages
The proliferation of social media has rendered it a critical arena for online marketing strategies. To optimize conversion rates, the landing pages must effectively respond to a visitor segment's pain points that they need solutions for. A one-size-fits-all approach is inadequate since even if the product meets the needs of all consumers, their priorities may diverge. In this study, we propose a pipeline for creating personalized landing pages that dynamically cater to visiting customers' specific concerns. As a use case, a pipeline will be utilized to create a personalized pharmacy discount card landing page, serving for the particular needs of chronic diabetics users seeking to purchase needed medications at a reduced cost. The proposed pipeline incorporates additional stages to augment the traditional online marketing funnel including acquisition of salient tweets, filtration of irrelevant ones, extraction of themes from relevant tweets, and generation of coherent paragraphs. To collect relevant tweets and reduce bias, Facebook groups and pages relevant to individuals with diabetes were leveraged. Pre-trained models such as BERT and RoBERTa were used to cluster the tweets based on their similarities. GuidedLDA exhibited superior performance in identifying customer priorities. Human evaluations revealed that personalized landing pages were more effective in getting the attention, building attraction by addressing their concerns and engaging the audiences. The proposed methodology offers a practical architecture for developing customized landing pages considering visiting customers' profiles and needs.