Tunazzina Islam

Tunazzina Islam

Ph.D. in Computer Science, Purdue University, West Lafayette.

Posts by Collection

portfolio

publications

Ex-Twit: Explainable Twitter Mining on Health Data

Published in The 7th International Workshop on Natural Language Processing for Social Media (SocialNLP 2019) In conjunction with 28th International Joint Conference on Artificial Intelligence (IJCAI 2019), 1900

[arXiv] [PDF] [Slide]

Recommended citation: Tunazzina Islam. 7th International Workshop on Natural Language Processing for Social Media (SocialNLP 2019) @ IJCAI-2019

Climate Claims Across Platforms: Comparing Meta Advertising and Bluesky Public Discourse

Published in Preprint 2026, 1900

Abstract

Climate communication reaches online audiences through distinct communication environments, including sponsored advertising and public user-generated discussion. Yet most cross-platform analyses focus on themes or stance rather than the specific claims communicators advance. We analyze 16,157 climate-related Meta advertisements and 18,781 Bluesky posts to examine how climate claims differ across these two online environments. Using a Toulmin-inspired constrained large language model (LLM) pipeline, we extract and validate explicit claims and compare them through topic, named-entity, and LIWC-based linguistic analyses. We find that Meta advertisements more often emphasize solution-oriented, locally targeted, and campaign-style claims, whereas Bluesky posts are more political, event-driven, and issue-centered. Meta claims also more frequently foreground clean energy, U.S. geographic contexts, and institutional actors, while Bluesky claims more often center on broad climate issues, fossil-fuel responsibility, and political actors. Linguistically, Meta claims exhibit more analytic and authoritative language, whereas Bluesky claims are more personal and expressive. These findings characterize observed differences between two structurally distinct climate communication environments and demonstrate how claim-level analysis complements conventional topic- and stance-based approaches by revealing not only what online discourse discusses, but also the claims communicators advance.

Recommended citation: Tunazzina Islam, Samantha Sudhoff, Jingying Hu, Cheng Wang, Zhaoqing Wu, Edward Wang. Under Review.

Analysis of Climate Campaigns on Social Media using Bayesian Model Averaging

Published in 6th AAAI/ACM Conference on AI, Ethics, and Society 2023 ([AIES-2023](https://www.aies-conference.com/2023/)), 1900

[Paper link] [arXiv] [Slide]

Abstract

Climate change is the defining issue of our time, and we are at a defining moment. Various interest groups, social movement organizations, and individuals engage in collective action on this issue on social media. In addition, issue advocacy campaigns on social media often arise in response to ongoing societal concerns, especially those faced by energy industries. Our goal in this paper is to analyze how those industries, their advocacy group, and climate advocacy group use social media to influence the narrative on climate change. In this work, we propose a minimally supervised model soup [57] approach combined with messaging themes to identify the stances of climate ads on Facebook. Finally, we release our stance dataset, model, and set of themes related to climate campaigns for future work on opinion mining and the automatic detection of climate change stances.

Recommended citation: Tunazzina Islam, Ruqi Zhang, Dan Goldwasser. 6th AAAI/ACM Conference on AI, Ethics, and Society 2023 (AIES-2023).

A Holistic Framework for Analyzing the COVID-19 Vaccine Debate

Published in Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies ([NAACL 2022](https://2022.naacl.org/)), 1900

[Paper link] [arXiv] [Slide]

Recommended citation: Maria Leonor Pacheco*, Tunazzina Islam*, Monal Mahajan, Andrey Shor, Ming Yin, Lyle Ungar, Dan Goldwasser.Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL 2022), 5821–583.

Understanding Microtargeting Pattern on Social Media

Published in 30th AAAI/SIGAI Doctoral Consortium ([AAAI-25 Doctoral Consortium](https://aaai.org/conference/aaai/aaai-25/doctoral-consortium-call/)), 1900

🏆 Won the best poster award in 2025 AAAI/SIGAI Doctoral Consortium

Recommended citation: Tunazzina Islam. AAAI-25 Doctoral Consortium.

When Objectives and Evaluators Disagree: Evaluating Transfer for Post-hoc Demographic Framing Mitigation

Published in Preprint 2026, 1900

Abstract

Post-hoc rewriting is increasingly used to mitigate disparities in LLM-generated text when the source model is inaccessible. However, a rewrite may appear successful because it improves the objective used for optimization or satisfies the evaluator used for selection, without transferring to a separately measured outcome. We study this problem using 660 gender pairs and 1,980 age contrasts of demographic-conditioned climate messages generated by three source models. We distinguish a deterministic, message-level Persuasion Bias Index (PBI) from two pairwise LLM-judged framing outcomes: an equal-weight mean over assertiveness, certainty, directive force, and warmth, and a separate direct overall-disparity rating. To test evaluator transfer, we separate the judge used for candidate selection from a frozen independent judge that evaluates blinded policy outputs without demographic, policy, selection-score, or acceptance metadata. A fixed policy that always edits the lower-PBI message transfers inconsistently across demographic axes and outcome aggregations. We therefore propose Adaptive-single rewriting, which generates candidates in both edit directions and selects at most one admissible rewrite using a separate selection judge. Under frozen independent evaluation, Adaptive-single significantly outperforms the controlled Lower-PBI policy on both framing outcomes for gender and age. A Both-sides policy yields the largest reduction in the equal-weight dimension mean but does not consistently improve the direct overall rating relative to Adaptive-single. Preservation audits using lexical similarity, embeddings, bidirectional natural language inference (NLI), and qualitative analysis reveal nontrivial intervention–preservation trade-offs. Our results show that post-hoc mitigation should be evaluated as transfer across objectives, evaluators, and editing policies, rather than from selection scores alone.

Recommended citation: Tunazzina Islam. Under Review.

Iterative Topic Taxonomy Induction with LLMs: A Case Study of Electoral Advertising

Published in In Findings of the 5th Asia-Pacific Chapter of the Association for Computational Linguistics & the 15th International Joint Conference on Natural Language Processing (AACL-IJCNLP 2026), 1900

[arXiv]

Recommended citation: Alexander Brady*, Tunazzina Islam*. In Findings of the 5th Asia-Pacific Chapter of the Association for Computational Linguistics & the 15th International Joint Conference on Natural Language Processing (AACL-IJCNLP 2026).

talks

teaching