IS Poster Day provides a forum for students across all of our graduate programs to showcase their ongoing or completed research, exchange research ideas and sharpen their presentation skills while competing for awards. The competition accepts submissions on a broad range of topics relevant to the field of information systems. Graduate student research is a very important part of our program as it prepares students for future careers and helps build problem-solving skills.
Below are the participants and award winners from our IS Poster Day that took place virtually on April 30th, 2021. Congratulations contest winners.
Ph.D. Completed Research:
1st Place: Rashidul Islam
Title: Debiasing Career Recommendations with Neural Fair Collaborative Filtering
Abstract: A growing proportion of human interactions are digitized on social media platforms and subjected to algorithmic decision-making, and it has become increasingly important to ensure fair treatment from these algorithms. In this work, we investigate gender bias in collaborative-filtering recommender systems trained on social media data. We develop neural fair collaborative filtering (NFCF), a practical framework for mitigating gender bias in recommending career-related sensitive items (e.g. jobs, academic concentrations, or courses of study) using a pre-training and fine-tuning approach to neural collaborative filtering, augmented with bias correction techniques. We show the utility of our methods for gender de-biased career and college major recommendations on the MovieLens dataset and a Facebook dataset, respectively, and achieve better performance and fairer behavior than several state-of-the-art models.
Tied for 2nd Place: Md Mahmudur Rahman
Title: DeepPseudo: Pseudo Value Based Deep Learning Models for Competing Risk Analysis
Abstract: Competing Risk Analysis (CRA) aims at the correct estimation of the marginal probability of occurrence of an event in the presence of competing events. Many of the statistical approaches developed for CRA are limited by strong assumptions about the underlying stochastic processes. To overcome these issues and to handle censoring, machine learning approaches for CRA have designed specialized cost functions. However, these approaches are not generalizable and are computationally expensive. This paper formulates CRA as a cause-specific regression problem and proposes DeepPseudo models, which use simple and effective feed-forward deep neural networks, to predict the cumulative incidence function (CIF) using Aalen-Johansen estimator-based pseudo values. DeepPseudo models capture the time-varying covariate effect on CIF while handling the censored observations. We show how DeepPseudo models can address covariate dependent censoring by using modified pseudo values. Experiments on real and synthetic datasets demonstrate that our proposed models obtain promising and statistically significant results compared to the state-of-the-art CRA approaches. Furthermore, we show that explainable methods such as Layer-wise Relevance Propagation can be used to interpret the predictions of our DeepPseudo models.
Tied for 2nd Place: Argho Sarkar
Title: VQA-AID: Visual Question Answering for Post-Disaster Damage Assessment and Analysis
Abstract: This research focuses on developing and implementing Visual Question Answering (VQA) tasks for post-disaster damage assessment purposes based on UAV imagery. Proper assessment and response based on assessment after any disaster can save many lives which are invaluable. Any response team needs to understand the scenario before starting their moves. Numerous analyses on a different source of data have been considered for damage assessment purposes. Unmanned Aerial Vehicle is one of the trusted ways of capturing image data from the affected area. These data contain essential information that needs to be considered for damage management purposes. Many traditional computer vision tasks such as classification, semantic segmentation, etc have been considered on these images to understand the images. However, our belief is that the visual question answering task will be one of the finest tasks that can be considered for post-disaster damage assessment purposes. Visual question answering task is so interactive that one can get the information by asking a question in natural language from an image. VQA system is able to provide high-level query-based information which is very difficult to obtain from other tasks. In this research, we present the VQA dataset, based on our FloodNet dataset, for post-disaster damage assessment and propose a supervised attention-based VQA model for this task.
Honorable Mention: Akiri Surely
Title: Design and Evaluation of a Team Stress Reflection System for Paramedic Trainees
Abstract: Stress is a leading cause of errors and mental health issues for Emergency Medical Service providers. Systems aimed to increase trainees’ awareness of their own stress reactions and impact on teammates may have long-term advantages for reducing stress in the field. We designed the Team Stress Reflection system as a proof-of-concept educational tool for paramedic training. The web-based interactive system aims to foster and promote stress reflection and awareness by presenting individuals and teams with electrodermal activity data and situated video recorded simulations. We conducted a user study to evaluate the system and trainees’ understanding of the data. The results indicate that interacting with the system promoted reflection on previous stress experiences and identified team dynamics as causes of potential stress. We discuss implications of the findings for team-based stress comprehension.
Ph.D. Late-breaking Results:
1st Place: Abdullah Aldaeej
Title: How Technical Debt Could Lead Startups to a High-risk of Failure
Abstract: Technical Debt (TD) refers to suboptimal technical solutions for expediting the software development in the short term, but entails extra work in future. It has been used as an excellent vehicle for startup organizations to cope with the limited resources and the uncertainty about the product-market fit. Despite the many advantages of TD for startups, it could have some negative impacts on internal software quality, and in some cases, these impacts lead to other external issues that are visible to customers. In this study, I conducted a multiple case study of two web and mobile app startups, retrospectively from the founding time until the study time. The primary data were collected via semi-structured interviews, by interviewing the technical founder (CTO) and 3-4 developers in each case. The interview data were complemented with secondary data which were collected from public documents about each case (e.g., mobile app online reviews, release comments, startup website and social accounts, etc.). I analyzed the data using thematic analysis to identify and categorize themes related to the TD effects, their time dimensions, and the relationship between the effects. We found two scenarios that illustrate how the effects of TD can be exacerbated to result in customer dissatisfaction and churn.
2nd Place: Vasundhara Misal
Title: Physiological Synchrony, Stress and Communication of Paramedic Trainees During Emergency Response Training
Abstract: Paramedics play a critical role in society and face many high stress situations in their day-to-day work. Lack of stress management can have short-term and long-term effects on paramedics such as effects on decision-making and cognitive skills, and mental health issues such as anxiety, depression and post-traumatic stress disorder (PTSD). As paramedics often work with a partner, the stress experienced by one partner may influence the stress of the other. This effect can be determined by assessing the team’s physiological synchrony. Physiological synchrony – the unconscious, dynamic linking of physiological responses such as electrodermal activity (EDA) – have been linked to stress and team coordination. In this preliminary analysis, we examined the relationship between EDA synchrony, perceived stress and communication between paramedic trainee pairs in the high-stake situation of emergency response training, as opposed to social situations or highly controlled lab settings. We used quantitative methods to find the relationship between physiological synchrony and stress, and between communication and physiological synchrony. We also qualitatively analyzed the communication to find the relation between communication and low and high synchrony segments. We found a moderate relationship indicating that higher levels of synchrony were correlated with lower perceived stress. Our initial results indicated a correlation between high physiological synchrony and social coordination and group processes. Moreover, communication between paramedic dyads was inversely related to physiological synchrony, i.e., communication increased during low synchrony segments of the interaction and decreased during high synchrony segments.
Honorable Mention: Munshi Mahbubur Rahman
Title: End-to-End Joint Modeling for Fake News Detection
Abstract: The rapid spread of misinformation, including misleading and manipulative content, is a current and urgent threat to our society and to our democracy. Over the past few years, there have been proposed a broad array of techniques to address this problem. In this work, we develop a deep learning-based end-to-end framework to detect fake news on social media which leverages BERT representations for the input sequences constructed with claim-evidence sentence pairs. Our method shows satisfactory performance in terms of the preliminary results based on a general fact-checking website (Snopes), a political fact-checking website (Politifact). We also intend to extend our approach using an attention mechanism that boosts the impact of important evidence sentences, while fading out the irrelevant ones. In future work, we also plan to compare our approach with state-of-the-art models in this domain and expect to achieve better performance due to the elegant properties, i.e. fine-tunable representations for claim-evidence sequences, augmented with attention mechanisms, of the proposed framework.
M.S. Completed Research:
1st Place: Anirudh Nagraj
Title: Investigating the navigational habits of individuals who are blind in India
Abstract: Assistive navigational technologies offer considerable promise to people with visual impairments. However, uptake of these technologies has traditionally been lower in low and middle-income countries (LMICs). In this paper, I describe a qualitative study undertaken with 14 people who identify as legally blind in an LMIC (India) to understand their requirements, experiences, and strategies undertaken when navigating with and without technology. I highlight key nuances that impact navigational habits, including strategies to navigate within busy urban environments, strategies to address the impact of the rainy season, techniques used to navigate at night, and dealing with the impact of limited infrastructure. Findings have led to the development of guidance to better support interface designers when designing more inclusive navigational solutions for a broader audience.
Audience Choice Awards:
4-way Tie for Ph.D. Completed Research:
Tashnim Chowdhury
Title: Self Attention Based Semantic Segmentation on A Natural Disaster Dataset
Abstract: Global image dependencies help in full image understanding. Self-attention based methods can map the mutual relationship and dependencies among pixels of an image and thus improve semantic segmentation accuracy. In this work, we propose two segmentation networks based on a novel baseline self-attention network. Compared to existing self-attention methods we utilize lower level feature maps to generate position attention modules which constitute a baseline network. This baseline network is incorporated with global average pooling and U-Net to create two segmentation schemes. These two segmentation networks are evaluated on a natural disaster dataset and perform excellent in damage assessment with a Mean IoU score of 95:61%.
Kamrun Naher Keya
Title: Equitable Allocation of Healthcare Resources with Fair Survival Models
Abstract: Healthcare programs such as Medicaid provide crucial services to vulnerable populations, but due to limited resources, many of the individuals who need these services the most languish on waiting lists. Survival models, e.g. the Cox proportional hazards model, can potentially improve this situation by predicting individuals’ levels of need, which can then be used to prioritize the waiting lists. Providing care to those in need can prevent institutionalization for those individuals, which both improves quality of life and reduces overall costs. While the benefits of such an approach are clear, care must be taken to ensure that the prioritization process is fair, and does not reinforce harmful systemic bias. We develop multiple fairness definitions and corresponding fair learning algorithms for survival models to ensure equitable allocation of healthcare resources. We demonstrate the utility of our methods in terms of fairness and predictive accuracy on three publicly available survival datasets.
Md Mahmudur Rahman
Title: DeepPseudo: Pseudo Value Based Deep Learning Models for Competing Risk Analysis
Abstract: Competing Risk Analysis (CRA) aims at the correct estimation of the marginal probability of occurrence of an event in the presence of competing events. Many of the statistical approaches developed for CRA are limited by strong assumptions about the underlying stochastic processes. To overcome these issues and to handle censoring, machine learning approaches for CRA have designed specialized cost functions. However, these approaches are not generalizable and are computationally expensive. This paper formulates CRA as a cause-specific regression problem and proposes DeepPseudo models, which use simple and effective feed-forward deep neural networks, to predict the cumulative incidence function (CIF) using Aalen-Johansen estimator-based pseudo values. DeepPseudo models capture the time-varying covariate effect on CIF while handling the censored observations. We show how DeepPseudo models can address covariate dependent censoring by using modified pseudo values. Experiments on real and synthetic datasets demonstrate that our proposed models obtain promising and statistically significant results compared to the state-of-the-art CRA approaches. Furthermore, we show that explainable methods such as Layer-wise Relevance Propagation can be used to interpret the predictions of our DeepPseudo models.
Redwan Walid
Title: Cloud-based Encrypted EHR System with Semantically Rich Access Control and Searchable Encryption
Abstract: Cloud-based electronic health records (EHR) systems provide important security controls by encrypting patient data. However, these records cannot be queried without decrypting the entire record. This incurs a huge amount of burden in network bandwidth and the client-side computation. As the volume of cloud-based EHRs reaches Big Data levels, it is essential to search over these encrypted patient records without decrypting them to ensure that the medical caregivers can efficiently access the EHRs. This is especially critical if the caregivers have access to only certain sections of the patient EHR and should not decrypt the whole record. In this paper, we present our novel approach that facilitates searchable encryption of large EHR systems using Attribute-based Encryption (ABE) and multi-keyword search techniques. Our framework outsources key search features to the cloud side. This way, our system can perform keyword searches on encrypted data with significantly reduced costs of network bandwidth and client-side computation.
Audience Choice Awards:
2-Way Tie for Ph.D. Late-breaking Results:
Iman Alshathri
Title: Applying virtual reality to motivate and prepare international students for community engagement
Abstract: There are numerous benefits to students participating in community engagement activities. However, levels of participation among international students are lower compared to their domestic counterparts. Virtual reality has been shown to have the potential to encourage behavioral change among individuals in a range of contexts. Researchers have yet to examine the impact of virtual reality to promote community engagement among international students. Our preliminary investigations suggest that this technology offers promise to motivate and prepare international graduate students to volunteer, by supporting the changing of perceptions towards volunteering. More specifically, this research aims to examine the role virtual reality technology plays in encouraging international students to volunteer in a community that is different from their own. Three studies are proposed as part of this work. Findings will lead to the development of suggestions to support community engagement trainers who are interested in applying virtual reality to motivate and prepare individuals to volunteer.
The first study targets experts in the community engagement (CE) field. The main goal of this phase is to ask community engagement practitioners about how international students are recruited and prepared for community engagement, and how to apply VR to prepare and motivate international students to participate in community engagement activities. The second study targets international students from different backgrounds and different age groups. The goal of this phase is to understand the impact of VR technology on international students’ attitudes toward community engagement activities. After Study II has been completed, a follow-up with experts in the community engagement field will be conducted through a participatory or co-design process. The goal is to fully understand how we develop an outline of a curricular unit that would provide a plan for integrating VR with classes to motivate and prepare international students for community engagement.
Munshi Mahbubur Rahman
Title: End-to-End Joint Modeling for Fake News Detection
Abstract: The rapid spread of misinformation, including misleading and manipulative content, is a current and urgent threat to our society and to our democracy. Over the past few years, there have been proposed a broad array of techniques to address this problem. In this work, we develop a deep learning-based end-to-end framework to detect fake news on social media which leverages BERT representations for the input sequences constructed with claim-evidence sentence pairs. Our method shows satisfactory performance in terms of the preliminary results based on a general fact-checking website (Snopes), a political fact-checking website (Politifact). We also intend to extend our approach using an attention mechanism that boosts the impact of important evidence sentences, while fading out the irrelevant ones. In future work, we also plan to compare our approach with state-of-the-art models in this domain and expect to achieve better performance due to the elegant properties, i.e. fine-tunable representations for claim-evidence sequences, augmented with attention mechanisms, of the proposed framework.