Research Area: Large Language Models (LLMs), Data Science, Human-Computer Interaction
Effect of Contextual Web Search Results on the Perceived Accuracy of LLM Hallucinations: I led a research study on the effect of search results on the perceived accuracy of LLM hallucinations, where we performed an online experiment (N = 560) that investigated how the provision of search results, either static (i.e., fixed search results provided by LLM) or dynamic (i.e., participant-led searches), affects participants’ perceived accuracy of LLM-generated content (i.e., genuine, minor hallucination, major hallucination), self-confidence in accuracy ratings, as well as their overall evaluation of the LLM, as compared to the control condition (i.e., no search results). This work was published in the journal "Computers in Human Behavior" in 2025.
Effect of Warning on Human Perception of and Engagement with LLM Hallucinations: I led this research on the efficacy of warnings on the perceived accuracy of and user engagement with LLM hallucinations by systematically varying the degree of hallucination (genuine, minor hallucination, major hallucination) and examining its interaction with warnings (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. This work was published in the First Conference on Language Modeling (COLM) in 2024.
Generative-AI Policies of Computer Science Conferences: As the use of Generative AI (Gen-AI) in scholarly writing and peer reviews continues to rise, it has become essential for the computing field to establish and adopt clear Gen-AI policies. In this study, we examined the landscape of Gen-AI policies across 64 major Computer Science conferences and offered recommendations for promoting more effective and responsible use of Gen-AI in the field. This work was published in the journal "Communications of the ACM (CACM)" in 2025.
Analyzing Text Segments Authored by Humans and LLMs to Uncover Nuanced Distinctions: I collaborated in a study where we investigated which text segment (introduction, body, or conclusion) differs most between human and LLM-generated text, inspired by a chess game analogy. This work has been accepted to EMNLP 2025 and is currently available on arXiv.
Task-Oriented Dialogue Summarization of User-Conversational Agent Interactions Using Chain-of-Interactions: I collaborated in a study where we examined the efficacy of LLM-supported dialogue summarization of user-agent conversations using chain-of-interactions prompting via computational and human-subjects studies. This work has been accepted to the Findings of EMNLP 2025.
LLM-Supported EFL Learning for Young Bangladeshi Learners: This is an ongoing collaborative study with a research group from CSE, BUET, Bangladesh which aims to observe the effects of LLM-supported English as a Foreign Language (EFL) learning mobile application on young Bangladeshi learners alongside the subjective experiences of the students, parents, and teachers.
Understanding the Dynamics of Online Study Groups: I conducted this study as part of my Ph.D. qualifying exam. This is a qualitative research study investigating the factors that were driving the growing popularity of online study groups. Through a series of interviews, I sought to identify the unique attributes that sustain member engagement within these communities.
Understanding Parents’ Struggle in Diagnosing Children with ASD: I collaborated in a survey study with parents of children who had Autism Spectrum Disorder (ASD) to understand parents’ struggles during the diagnosis process.
Research Area: Machine Learning, Speech Processing, Affective Computing
Anger Detection from Bengali and English Audio-Textual Conversations: In this work, we proposed novel deep learning-based approaches for both offline and online anger detection from audio-textual data obtained from real-life conversations. For offline anger detection, which detects the anger of a given audio-textual conversation, we introduced an ensemble approach that adapts attention-based CNN architecture, gender classifier, and BERT-based textual features to derive the anger of a conversion. On the other hand, for online anger detection, which predicts anger in the conversation of the subsequent timestamps from the conversations of previous timestamps, we proposed a transformer-based audio and textual ensemble technique to predict the anger of a future conversation. We demonstrate the efficacy of our proposed approaches using two datasets: the Bengali call center dataset and the IEMOCAP dataset. This work appeared in the Proceedings of the Affective Computing and Intelligent Interaction (ACII) Conference, in 2022.
Research Area: Spatial Databases
Trajectories that Provide the Best Visibility of Target Objects: The advancements in large-scale 3D modeling inspired applications that combined visibility and spatial queries, which in turn could be integrated with user trajectories to provide answers for many real-life user queries, such as “How can I choose the route which provides the best view of a historic site?”. In this work, we proposed and investigated the k Aggregate Maximum Visibility Trajectory (kAMVT) query and its variants. Given sets of targets, obstacles, and trajectories, the kAMVT query finds top-k trajectories that provide the best view of the targets. To provide an efficient solution to our problem, we employed obstacle and trajectory pruning mechanisms. In addition, we conducted extensive experimental studies using large synthetic and real datasets. This work appeared in Geoinformatica, Springer, in 2019.