Prof. Bo An is a President’s Chair Professor in Computer Science, Head of Division of Artificial Intelligence, and Co-Director of Artificial Intelligence Research Institute (AI.R) at Nanyang Technological University, Singapore. He received the Ph.D degree in Computer Science from the University of Massachusetts, Amherst. His current research interests include artificial intelligence, multiagent systems, computational game theory, reinforcement learning, and optimization. His research results have been successfully applied to many domains including infrastructure security, sustainability and e-commerce. He has published over 150 referred papers at AAMAS, IJCAI, AAAI, ICLR, NeurIPS, ICML, AISTATS, ICAPS, KDD, UAI, EC, WWW, JAAMAS, and AIJ. Dr. An was the recipient of the 2010 IFAAMAS Victor Lesser Distinguished Dissertation Award, an Operational Excellence Award from the Commander, First Coast Guard District of the United States, the 2012 INFORMS Daniel H. Wagner Prize for Excellence in Operations Research Practice, 2018 Nanyang Research Award (Young Investigator), and 2022 Nanyang Research Award. His publications won the Best Innovative Application Paper Award at AAMAS’12, the Innovative Application Award at IAAI’16, and the best paper award at DAI’20. He was invited to give Early Career Spotlight talk at IJCAI’17. He led the team HogRider which won the 2017 Microsoft Collaborative AI Challenge. He was named to IEEE Intelligent Systems' "AI's 10 to Watch" list for 2018. He was PC Co-Chair of AAMAS’20 and General Co-Chair of AAMAS’23. He will be PC Chair of IJCAI’27. He is a member of the editorial board of JAIR and is the Associate Editor of AIJ, JAAMAS, IEEE Intelligent Systems, ACM TAAS, and ACM TIST. He was elected to the board of directors of IFAAMAS, senior member of AAAI, and Distinguished member of ACM.
Keynote talk: Deep Reinforcement Learning for Quantitative Trading
In the last decade, we have witnessed a significant development of AI-powered quantitative trading (QT), due to its instant and accurate order execution, and capability of analyzing and processing large amount of data related to the financial market. Traditional AI-powered QT methods discover trading opportunities based on either heuristic rules or financial prediction. However, due to the high volatility and noisy nature of financial market, their performance is not stable and highly reply on the market condition. Recently, reinforcement learning (RL) becomes an appealing approach for QT tasks owing to its stellar performance on solving complex decision-making problems. This talk will discuss some recent research progress in RL for QT and future directions. More information is at http://trademaster.ai/ and https://github.com/TradeMaster-NTU/TradeMaster.
Dr. Chung-Chi Chen is currently a researcher at the Artificial Intelligence Research Center, AIST, Japan. His scholarly pursuits revolve around the intricate realm of financial opinion mining and the nuanced understanding and generation of financial documents. He has orchestrated the FinNLP/FinWeb workshop series within prestigious conferences such as IJCAI, WWW, EMNLP, and IJCNLP-AACL since 2019. He has guided the FinNum and FinArg shared task series on the NTCIR from 2018 to 2023. He was also a presenter in the AACL-2020, EMNLP-2021, and ECAI-2024 tutorials. He served as Program Co-Chair of NTCIR-18, Senior Area Chair of ACL-2024, and PC member in many representative conferences. In academic competitions, he fortunately won 2nd place in the SIGIR Early Career Researcher Award, in addition to two Thesis Awards and Technology Innovation Award. Beyond academia, he has also ventured into the dynamic realm of FinTech. He earned one prize in a startup competition and four prizes in FinTech competitions. In addition to FinTech, he also has been honored with two prizes in LegalTech competitions.
Keynote talk: Financial Opinion Recommendation from Personalization, Profitability, and Rationality Aspects
In the era of rapid Internet and social media platform development, individuals readily share their viewpoints online. The overwhelming quantity of these posts renders comprehensive analysis impractical. This necessitates an efficient recommendation system to filter and present significant, relevant opinions. The recommender system in this scenario could be separated into three categories: based on the investor's personal interest, profitability, or rationality. In this talk, we will share our pilot explorations [1,2,3,4] on these three aspects and highlight the importance of adopting argument mining and identifying professionalism in financial opinion recommendations.
[1] Evaluating the Rationales of Amateur Investors. In Proceedings of The Web Conference 2021 (WWW'21)
[2] Personalized Dynamic Recommender System for Investors. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR'23)
[3] Argument-Based Sentiment Analysis on Forward-Looking Statements. In Findings of the Association for Computational Linguistics: ACL 2024
[4] Professionalism-Aware Pre-Finetuning for Profitability Ranking. In Proceedings of The 33rd ACM International Conference on Information and Knowledge Management (CIKM'24)
Dr. Seongho Eun is a Senior Data Scientist at Hana Institute of Technology, Hana TI. He received a Ph.D. and M.S. degree in Management Engineering and a B.S. degree in Mathematical Sciences, all from Korea Advanced Institute of Science and Technology. He researched the impact of IT technology on the economy during doctoral course. His research interests include application of machine learning methods, credit rating, and personalized credit line to banking industry.
Invited talk: Leveraging Internal Transaction Data for Personalized Loan Products: A Methodology for Broadening Clientele and Increasing Bank Loyalty
As banks strive to bolster profitability, there arises a pressing need to broaden the clientele for loan products. This study presents a methodology for designing loan products based on transaction performance using internal bank data. Targeting customers with robust transaction histories but limited access to credit loans, we develop a loan product tailored to their needs. Leveraging two years of credit approval data and machine learning techniques including XGBoost and LightGBM regression, loan amounts are predicted for all bank customers. The study highlights the potential of leveraging internal transaction data to expand the profitable customer base and increase brand loyalty.
Dr. Sang Hyun Park is a Senior Data Scientist at Hana Institute of Technology, Hana TI. Prior to his current position, he worked as a Data Scientist at the Marketing Team of Memory division, Samsung Electronics. He received a Ph.D. degree in Management Engineering and an M.S. and a B.S. degree in Mathematical Sciences, all from Korea Advanced Institute of Science and Technology. His research interests include application of data-driven methods and unsupervised anomaly detection to financial fraud detection.
Invited talk: Real-time phishing app detection using real-world data in Korea
In South Korea, telecommunications financial fraud such as smishing and voice phishing is getting more and more sophisticated in recent years. A real-time, automated system for malicious (phishing) app detection is required for Korean banks to prevent damage on customers without excessive human effort. To this end, we test machine learning models such as gradient-boosted decision trees and random forests on android application features and information on customers of Hana Bank. Evaluating models with 92 daily sets of data (ten-week train data and one-day test data in each set), we find that applications' package names contain meaningful information for detecting phishing apps, among others. In addition, while interpretable statistical features of package names (e.g., length or proportion-based measures) have considerable roles in model performance, adding extracted features using work embedding models such as fastText further improves detection performance. We also find that the default threshold 0.5 yields better performance compared to a daily-adjusted thresholding scheme using validation data.
Namhyoung Kim graduated from Hanyang University in 2020 with a degree in Economics & Finance. Since 2021, he has been pursuing a Ph.D. in Industrial Engineering at Hanyang University. His research interests are centered on factor models, financial representation, and the application of artificial intelligence in finance.
Title: Risk Propensity-specific Portfolio Recommendation via Self-supervised Learning
This paper introduces a two-stage framework to create possible investment strategies that can be tailored to investors, particularly in risk tolerance, which profoundly influences financial decisions and potential returns. The framework initially employs deep online clustering based on contrastive learning to identify and align investment assets with risk profiles. Subsequently, the security selection is carried out based on momentum bimodality. In particular, we propose FactorMAE (Bimodal factor model using a Momentum Auto-Encoder), a novel algorithm designed for selecting securities within each risk cluster based on momentum, under the assumption that well-performing assets will sustain their trajectory. Unlike conventional models that are reliant on historical returns and volatility, FactorMAE capitalizes on current market trends, thereby enhancing portfolio adaptability and potential returns in volatile market conditions. The results of the clustering and backtesting, conducted on various asset types traded in the Korean financial market, highlight clear distinctions between the risk levels of six risk clusters and demonstrate improved investment performance through security selection.
Dae-Young Park is currently working at the Financial Security Institute (FSI) in the AI technology team, focusing on data and AI security, utilization, and software engineering. He previously worked as a back-end developer at Entrepreple. Dae-Young is also a Ph.D. candidate in the Department of Computer Science at KAIST, where he earned his M.S. degree, following his B.S. in Computer Science and Engineering from Hanyang University. His goal is to conduct practical research in collaboration with academic experts in the financial sector.
Title: Graph-based Tabular Data Synthesis Model for More Effective Financial Application
Although there are existing tabular data synthesis models available for financial applications, there are still few studies considering two characteristics of financial fraud detection (FFD) data: (a) extreme class imbalance ratio and (b) high attribute ratio of non-normal distribution. We propose a novel graph-based tabular data synthesis model through combination of graph-based feature engineering, graph-theoretical analysis, and graph centrality indicators. Experimental results on three popular FFD benchmark datasets show that our method outperforms five different baseline methods.
Youngjin Park is a Ph.D. student at the Kim Jaechul Graduate School of Artificial Intelligence at KAIST, where he began his studies in 2023. His research interests lie in the areas of time series analysis, explainable AI (XAI), and reinforcement learning. Prior to starting his doctoral studies, Youngjin worked as a Reinforcement Learning Researcher at Netmarble AI Center from 2022 to 2023. He also has industry experience as a Quantitative Researcher at SQR, where he worked from 2021 to 2022. Youngjin holds a Master of Science degree in Computer Science and Engineering from UNIST, which he obtained in 2021. During his graduate studies, he co-authored two research papers: "Improved predictive deep temporal neural networks with trend filtering" published in the Proceedings of the First ACM International Conference on AI in Finance in 2020, "Analyzing Stock Price Change Explainable Sentence Classification Model using Shapley Value Estimation" published in KIISE Transactions on Computing Practices in 2020.
Title: Surrogate for Marginal Amortized SHAP for Explaining Recommendation Systems
Explainable AI (XAI) techniques are crucial for enhancing the transparency and interpretability of recommendation systems. The prominence of Shapley values in explaining model decisions necessitates efficient computational approaches, particularly for large datasets. In this study, we introduce SMASHAP (Surrogate for Marginal Amortized SHAP), a novel method that utilizes a surrogate model to provide single inference estimations from marginal distributions. We examine the use of SMASHAP to efficiently explain how a recommendation model works on both global and local levels. By aggregating the absolute SHAP values, we observe that some features have a high impact on the model's predictions. Furthermore, by comparing the ratio of absolute SHAP values between the entire dataset and instances where the model made incorrect predictions, we identify features that have relatively high SHAP values in those cases. To demonstrate the utility of SMASHAP for global explanations, we show that the proposed methodology is faster than existing methods.
Lubingzhi Guo is a current PhD student within the Financial Informatics theme at the University of Glasgow, researching the intersection of natural language processing for finance and large language models. Her core research topics include: financial entity and relationship extraction from financial text documents; financial knowledge graph construction; and graph embedding generation for companies. She works on a range of financial use-cases, focusing on financial asset recommendation and company report generation.
Title: Comparing the Impact of Financial Knowledge Graphs from Financial Reports and Wikidata in Asset Recommendation
Too often when we examine financial use-cases, we only use market data to represent the assets we recommend. However, it is also critical to model the companies that those assets represent. This involves understanding what a company does, the future plans of that company, as well as its fundamentals. In this work we examine the extent relational knowledge can fill this gap. In particular, we create two different types of financial knowledge graphs: one with general company information; and one generated from company annual reporting. We show that as we might expect, integrating company insights from these knowledge graphs enhance recommendation performance(y up-to 10% monthly RoI), but more interestingly the best way to embed each knowledge graph is different.
Prof. Richard McCreadie is a Senior Lecturer (UK equivalent to Associate Professor) in the areas of real-time Information Retrieval, machine learning, big data stream processing and evaluation methodologies over streaming data, currently based within the IR Group/Terrier Team of the School of Computing Science at the University of Glasgow. He specialises in real-time IR technologies that enable users to efficiently and effectively find information in big data streams; the research and development of automatic (assistive) agents to detect events, extract knowledge and summarize information within big data streams; as well as methodologies to evaluate systems that process such streams. He is the head of the Financial Informatics Research Theme at Glasgow, and is a steering committee member for the Text Retrieval Conference (TREC).
Title: FAR-Trans: An Investment Dataset for Financial Asset Recommendation
Financial asset recommendation (FAR) is a sub-domain of recommender systems which identifies useful financial securities for investors, with the expectation that they will invest capital on the recommended assets. FAR solutions analyse and learn from multiple data sources, including time series pricing data, customer profile information and expectations, as well as past investments. However, most models have been developed over proprietary datasets, making a comparison over a common benchmark impossible. We introduce an new dataset FAR-Trans, the first public dataset for FAR, containing pricing information and retail investor transactions acquired from a large European financial institution, as well as provide a bench-marking comparison between eleven FAR algorithms over the data for use as future baselines.
Yejin Kim is an M.Sc. student at the Financial Engineering Lab at Ulsan National Institute of Science and Technology (UNIST), where she previously earned her Bachelor’s degree in Industrial Engineering. She is researching AI-driven financial services, focusing on a stock recommendation service based on individual investors’ purchasing data. Additionally, she has developed AI solutions for financial services using recommendation systems, GNN, and NLP.
Title: Stock Recommendations for Individual Investors: A Temporal Graph Network Approach with Mean-variance Efficient Learning
Recommender systems could be helpful for individuals to make well-informed decisions in complex financial markets. While many studies have focused on predicting stock prices, even advanced models fall short of accurately forecasting them. Additionally, previous studies indicate that individual investors often disregard established investment theories, favoring their personal preferences instead. This presents a challenge for stock recommendation systems, which must not only provide strong investment performance but also respect these individual preferences. To create effective stock recommender systems, three critical elements must be incorporated: 1) individual preferences, 2) portfolio diversification, and 3) the temporal dynamics of the first two. In response, we propose a new model, Portfolio Temporal Graph Network Recommender PfoTGNRec, which can handle time-varying collaborative signals and incorporates diversification-enhancing sampling. On real-world individual trading data from the Greek stock market, our approach demonstrated superior performance compared to various baselines, including cutting-edge dynamic embedding models and existing stock recommendation models. Our model could improve investment performances while effectively capturing individual preferences.