We present a practical system that supports in-depth analysis of cryptocurrency markets through timeline-based event detection and contextual summarization. Our framework processes continuous news streams, identifies price-relevant events, and organizes them into semantic timelines with concise background summaries generated by large language models (LLMs). This design allows traders and analysts to retrospectively explore events alongside price charts, facilitating a deeper understanding of how news developments relate to market fluctuations. By transforming unstructured news data into structured insights, the system provides a valuable tool for market analysis, risk evaluation, and behavioral studies in volatile trading environments.

Recent advancements in large language models (LLMs) have enabled in-context learning (ICL)-based methods that significantly outperform fine-tuning approaches for text-to-SQL tasks. However, their performance is still considerably lower than that of human experts on benchmarks that include complex schemas and queries, such as BIRD. This study considers the sensitivity of LLMs to the prompts and introduces a novel approach that leverages multiple prompts to explore a broader search space for possible answers and effectively aggregate them. Specifically, we robustly refine the database schema through schema linking using multiple prompts. Thereafter, we generate various candidate SQL queries based on the refined schema and diverse prompts. Finally, the candidate queries are filtered based on their confidence scores, and the optimal query is obtained through a multiple-choice selection that is presented to the LLM. When evaluated on the BIRD and Spider benchmarks, the proposed method achieved execution accuracies of 65.5% and 89.6%, respectively, significantly outperforming previous ICL-based methods.

Minjin Choi, Sunkyung Lee, Eunseong Choi, Heesoo Park, Junhyuk Lee, Dongwon Lee, Jongwuk Lee. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Substance Use Disorder (SUD) is a devastating disease that leads to significant mental and behavioral impairments. Its negative effects damage the homeless youth population more severely (as compared to stably housed counterparts) because of their high-risk behaviors. To assist policymakers in devising effective and accurate long-term strategies to mitigate SUD, it is necessary to critically analyze environmental, psychological, and other factors associated with SUD among homeless youth. Unfortunately, there is no definitive data-driven study on analyzing factors associated with SUD among homeless youth. While there have been a few prior studies in the past, they (i) do not analyze variation in the associated factors for SUD with geographical heterogeneity in their studies; and (ii) only consider a few contributing factors to SUD in relatively small samples. This work aims to fill this gap by making the following three contributions: (i) we use a real-world dataset collected from ~1,400 homeless youth (across six American states) to build accurate Machine Learning (ML) models for predicting the susceptibility of homeless youth to SUD; (ii) we find a representative set of factors associated with SUD among this population by analyzing feature importance values associated with our ML models; and (iii) we investigate the effect of geographical heterogeneity on the factors associated with SUD. Our results show that our system using adaptively boosted decision trees achieves the best predictive accuracy out of several algorithms on the SUD prediction task, achieving an Area Under the ROC Curve of 0.85. Further, among other things, we also find that both Post-Traumatic Stress Disorder (PTSD) and depression are very strongly associated with SUD among homeless youth because of their propensity to self-medicate to alleviate stress. This work is done in collaboration with social work scientists, who are currently evaluating the results for potential future deployment.