paper 1: Understanding vs. Generation: LLMs Are Better at Comprehending Low-Resource Languages Like Urdu Than Generating Text
Taaha Saleem Bajwa
Abstract:
Large Language Models (LLMs) are predominantly trained on English data, leading to significant performance challenges for low-resource languages. This study focuses on Urdu as a case study to explore how LLMs process low-resource languages. We find that when prompted in a low-resource language, LLMs primarily reason internally in English, and this internal reasoning is more coherent than the generated text in the target language. This contrast highlights a gap between the model’s ability to comprehend low-resource languages and its ability to generate text in them. By analyzing these mechanisms, this work underscores the need for targeted improvements to enhance LLM performance for low-resource languages such as Urdu.
Paper 2: Optimizing Portfolio Performance through Clustering and Sharpe Ratio-Based Optimization: A Comparative Backtesting Approach
Keon Vin Park
Abstract:
Optimizing portfolio performance is a fundamental challenge in financial modeling, requiring the integration of advanced clustering techniques and data-driven optimization strategies. This paper introduces a comparative backtesting approach that combines clustering-based portfolio segmentation and Sharpe ratiobased optimization to enhance investment decision-making. First, we segment a diverse set of financial assets into clusters based on their historical log-returns using K-Means clustering. This segmentation enables the grouping of assets with similar return characteristics, facilitating targeted portfolio construction. Next, for each cluster, we apply a Sharpe ratio-based optimization model to derive optimal weights that maximize risk-adjusted returns. Unlike traditional mean-variance optimization, this approach directly incorporates the trade-off between returns and volatility, resulting in a more balanced allocation of resources within each cluster. The proposed framework is evaluated through a backtesting study using historical data spanning multiple asset classes. Optimized portfolios for each cluster are constructed and their cumulative returns are compared over time against a traditional equal-weighted benchmark portfolio. Our results demonstrate the effectiveness of combining clustering and Sharpe ratio-based optimization in identifying high-performing asset groups and allocating resources accordingly. The selected cluster portfolio achieved a total return of 140.98%, with an annualized return of 24.67%, significantly outperforming the benchmark portfolio, which achieved a total return of 107.59% and an annualized return of 20.09%. The Sharpe ratio of the selected portfolio (0.84) further highlights its superior risk-adjusted performance compared to the benchmark portfolio (0.73). This methodology provides a systematic and data-driven approach to portfolio optimization, achieving superior returns while maintaining a reasonable level of risk.
Paper 3: Assessing Youth Perceptions and Acceptance of AI Chatbots for Enhancing Sexual and Reproductive Health Literacy in Nigeria and Bangladesh
Habeeb Abdulrauf
Abstract:
Recent advancements in artificial intelligence (AI) have spurred the development of innovative digital tools that are reshaping health communication (Laranjo et al., 2018; Fitzpatrick, Darcy, & Vierhile, 2017). These tools offer considerable potential to enhance health literacy, particularly for populations facing persistent barriers to accessing sexual and reproductive health (SRH) information and services. In countries such as Nigeria and Bangladesh, adolescents and youth encounter multiple obstacles—including stigma, shyness, financial constraints, negative attitudes among health workers, poor service integration, lack of privacy, and restrictive religious and social norms (Afolabi & Afolabi, 2019; Aji et al., 2020; Ajibade et al., 2022; Nmadu, Mohamed, & Usman, 2020; Sultana & Subarna, 2018)—which impede their access to quality SRH information. In this context, AI chatbots present a promising alternative by offering confidential, free, immediate, and culturally sensitive access to vital SRH knowledge. Such digital interventions may bypass traditional barriers by enabling young people to seek guidance without fear of stigma or judgment (Hollis et al., 2017), thereby potentially mitigating adverse outcomes such as sexually transmitted infections (STIs), HIV, unintended pregnancies, and unsafe abortions.
However, the successful implementation of AI-driven health literacy strategies hinges on youth perceptions and acceptance of these technologies. Thus, it is imperative to assess whether young people in Nigeria and Bangladesh consider AI chatbots trustworthy, useful, and acceptable as a means of obtaining accurate SRH information. This study employs a mixed-methods design grounded in the Technology Acceptance Model (TAM) (Davis, 1989) and Hofstede’s Cultural Dimensions Theory (Hofstede, 2001) to examine how individual cognitive evaluations—specifically, perceived usefulness and ease of use—interact with broader cultural factors, such as traditional norms, stigma, and religious beliefs, to shape attitudes toward these digital interventions. The following research questions guide our investigation:
How do youth in Nigeria and Bangladesh perceive the use of AI chatbots for sexual and reproductive health information? To what extent do these perceptions influence the acceptance and intended use of AI chatbots for SRH information? What do youth identify as the primary benefits and drawbacks of utilizing AI chatbots for SRH information? Findings will be discussed within these theoretical frameworks to provide insights into the factors influencing the adoption of AI chatbots as a viable means of accessing SRH information. This research aims to inform the design of culturally sensitive digital health interventions and guide policymakers and health practitioners in overcoming barriers to SRH services in low-resource settings.
Paper 4: Evaluating the Feasibility of Universal Basic Income (UBI) in Bangladesh Using Multivariate Forecasting with XGBoost
Manash Sarker, Fahmida Rahman Liza, Julshan Alam Ratu, Md Saiful Islam, Abdullah Al Farooq
Abstract:
This study represents the first application of artificial intelligence (AI) techniques, specifically XGBoost, to evaluate the feasibility of Universal Basic Income (UBI) in Bangladesh, a country where over 20% of the population lives below the poverty line and wealth inequality remains high. Unlike previous studies that primarily focused on short-term socio-economic impacts or prioritized developed economies with abundant government revenue, this research addresses the unique challenges of a resource-constrained developing country. By applying multivariate forecasting with XGBoost, this study captures complex, non-linear relationships among economic indicators such as GDP growth, tax revenue, and government expenditure. The model achieved high predictive accuracy (90%), with a precision of 92%, and demonstrated strong performance through metrics like Root Mean Squared Error (RMSE) of 0.1338 and R2 score of 0.78. Using SHAP (SHapley Additive exPlanations), the study provided interpretable insights into the economic drivers influencing UBI feasibility, such as GDP growth and Debt-to-GDP ratios. Key contributions include: ● Developing a new dataset for UBI forecasting in resource-constrained settings. ● Breaking through the limitations of previous studies by offering a long-term, data-driven evaluation of UBI feasibility in Bangladesh. ● Addressing budgetary constraints unique to developing economies while considering economic sustainability. ● Providing actionable insights for policymakers to address socio-economic disparities. Despite its strong predictive performance, the study's dataset size limits it. Future work could address this limitation by incorporating cross-country comparisons, larger datasets, and advanced models like Temporal Fusion Transformers. This research offers a pioneering, AI-driven approach to addressing UBI feasibility, with significant implications for emerging regions facing poverty, inequality, and economic shocks.
Paper 5: Improving Sentiment Analysis of Low Resource Languages with LLMs and Strategic Prompting
Ibukunoluwa Folajimi
Abstract:
Sentiment analysis for low-resource languages is challenging due to limited data and resources. In this paper, we propose an approach that combines Large Language Models (LLMs) and prompt engineering to improve sentiment detection for Yoruba, a low resource language spoken in South-We. First, we fine-tuned a multilingual LLM using a small dataset of annotated texts. Next, we applied prompt engineering techniques to craft prompts that effectively guide the model’s understanding of context and sentiment. Data augmentation using back-translation was also utilized to expand the dataset. Our approach achieved an accuracy of 85% and an F1-score of 82%, outperforming traditional methods by 15%. These results highlight the potential of LLMs and prompt engineering to bridge language gaps and make AI tools more inclusive for emerging regions.
Paper 6: Towards Culturally Inclusive Knowledge Dissemination Using AI Agents
Mahfuz Ahmed Anik, Abdur Rahman, Azmine Toushik Wasi, Md Manjurul Ahsan
Abstract:
Artificial Intelligence plays a crucial role in knowledge dissemination but often reflects Western-centric biases, leading to a lack of inclusivity and fairness. This highlights the need for AI systems that address cultural disparities. To tackle this, we propose an agent-based AI framework focused on enhancing fairness, explainability, and participatory governance. Our multi-agent system includes agents for bias detection, user feedback integration, and transparency. Key contributions include a dynamic bias mitigation framework, continuous model improvement, and strategies to build trust in AI. Our approach ensures improved cultural inclusivity, transparency, and user trust, offering a path toward more equitable AI adoption in marginalized communities.
Paper 7: Regional Determinants of Domestic Violence in India: Machine Learning Approach
Greeshma Balabhadra, Mansi Sharma
Abstract:
Abstract: In this paper, we examine the factors affecting the likelihood of an Indian woman experiencing domestic violence. In addition, we also analyze how these factors vary across six regions in India. Using a nationally representative survey, we employ machine learning (ML) techniques such as logistic regression, Least Absolute Shrinkage and Selection Operator (LASSO), Random Forest (RF), Extreme Gradient Boosting (XGB), and SHapley Additive exPlanations (SHAP), along with dimension reduction methods like autoencoder. Among logistic regression, LASSO, and XGB, RF consistently outperforms the others in both the full sample and regional analyses. While different models prioritize features differently, key predictors of spousal abuse — husband’s control issues, woman’s age at first cohabitation, woman’s family background, and her physical stature — consistently emerge as significant both in the overall sample and in the subsamples
divided across the six regions in India. However, at the regional level, cluster analysis reveals significant intra-regional variation, reinforcing the need for localized interventions to effectively address domestic violence in India.
We are excited to announce a Special Call for Submissions into the AI in Emerging Regions, a special track at the Social Impact of AI: Research, Diversity, and Inclusion Frameworks (SIAI-ReDI) Workshop during the AAAI-25 Conference. This track aims to spotlight the innovative contributions of AI researchers, practitioners, and scholars from underrepresented regions or those whose work addresses the distinct challenges and opportunities in emerging regions. The track's primary focus is on tackling local challenges, unlocking new possibilities, and advancing AI adoption in these areas. By providing a dedicated platform to amplify diverse voices, foster meaningful global collaboration, and celebrate the transformative power of AI, this showcase aims to inspire and drive impactful change in emerging regions.
We invite authors to submit a 2-page extended abstract summarizing their AI research, including completed studies, research in progress, or innovative ideas. Submissions should highlight the work’s impact, relevance to the workshop’s themes, and its significance for addressing challenges or opportunities in emerging regions
Submission Deadline: February 7, 2025
Notification of Acceptance: February 14, 2025
Poster and Video Submission Deadline: February 21, 2025
Workshop Date and Time: March 1, 2025, 10:00 AM – 3:00 PM
Paper submission link: https://openreview.net/group?id=AAAI.org/2025/Workshop/SIAI-Redi_SPP
Full call for papers: https://www.siai.co/aier
This call is open to:
Researchers, students, and practitioners currently working in or studying AI in emerging regions.
Note: Emerging regions refer to areas, typically in developing countries or underrepresented communities, where access to technological resources, economic development, and research infrastructure is still growing but has immense potential for impactful innovation.
Contributions that directly address challenges or demonstrate impactful outcomes in these regions.
All accepted papers will be published in the workshop proceedings.
Accepted authors will be invited to submit a prerecorded video and poster describing their work
Posters and videos will be featured on the program website.
Best Poster Award will be presented to the most outstanding submission, recognizing exceptional quality and impact.
Submissions are invited on, but not limited to, the following themes:
AI and Marginalized Communities
AI for Social Good and Public Impact
AI in Global and Cross-Cultural Contexts
AI, Disability Inclusion, and Accessibility
AI, Ethics, and Human Rights
Algorithmic Bias, Fairness, and Ethical AI
Case Studies on AI’s Societal Impact (e.g., education, healthcare, law enforcement)
Collaborative Approaches to AI for Public Good
Cross-Cultural Perspectives on AI Governance and Accountability
Diversity, Equity, and Inclusion in AI Development
Educational Frameworks and AI Literacy for Inclusion
Frameworks for Evaluating AI’s Social Impact
Gender and Intersectionality in AI
Inclusive AI Development Processes
Mitigating Algorithmic Bias and Discrimination
Public Trust, Transparency, and Participatory AI Practices
Regulatory Frameworks and Policies for Inclusive AI
Transparent, Explainable, and Accountable AI Systems
Authors are invited to submit their 2-page extended abstract by February 7, 2025. Please structure your abstract using the following sections:
Title
Authors and Affiliations
Introduction
Objectives
Methodology
Key Findings or Contributions
Relevance to Emerging Regions (Explain the practical, societal, or economic implications of your work to emerging regions)
Future Directions
References
External Links (optional links to external libraries, code, dataset and other resources to support the work)
Formatting Requirements:
Authors must use AAAI 2025 authors' kit: https://aaai.org/authorkit25/ to format their documents
Registration:
All accepted authors should select the most suitable registration option at AAAI-25 registration link: https://aaai.org/conference/aaai/aaai-25/registration/
For inquiries, please email the organizers: folajimiy@wit.edu, deligiannidisl@wit.edu, othmans1@wit.edu
We look forward to your contributions and to celebrating the rich and diverse AI research from emerging regions!