A recent study has introduced an interactive 3D visualization platform designed to help financial advisors and institutional investors navigate the complexities of multi-objective portfolio optimization, particularly when integrating environmental, social, and governance (ESG) factors. The research addresses a significant challenge in modern finance: the need for transparency when balancing traditional financial metrics like risk and return with diverse, non-financial objectives. By utilizing multi-objective evolutionary algorithms (MOEAs) and providing a visual interface for exploring Pareto-optimal solutions, the system aims to make the trade-offs inherent in multi-objective decision-making explicit and understandable. This approach moves beyond simple two-dimensional representations, which often fall short in scenarios involving more than two competing objectives.
Financial portfolio selection is a complex process that must account for multiple factors, including expected returns, risk tolerance, and time horizons, while increasingly incorporating ESG considerations for sustainable and socially responsible investing. The lack of a globally standardized framework for ESG assessment has led to variability across investment practices, making it difficult for investors to align their portfolios with ethical and strategic values. The researchers developed their platform using a dataset of 119 popular mutual funds provided by Morningstar, Inc., which includes 15 distinct ESG factors. By using tracking error as a joint factor to represent risk and return, the researchers formulated a multi-objective optimization problem that seeks to maximize positive ESG scores while minimizing negative ones, all while maintaining the desired risk profile of the investor.
The interactive visualization platform consists of two main components: an interactive 3D visualization interface and a portfolio dashboard. The visualization interface allows users to select any three dimensions—such as tracking error, various ESG scores, or user-specified metrics—to observe the distribution of non-dominated portfolios in a 3D plot and its corresponding 2D projections. The system also integrates several methods to identify potential optimal solutions, including knee-point detection, which does not require predefined weight preferences, and multi-criteria decision-making (MCDM) methods that allow users to input their specific priorities. These features enable users to interactively explore the solution space, rotate plots, zoom in on specific data points, and visually compare how different portfolios stack up against predefined benchmarks.
The portfolio dashboard complements the visualization by offering a detailed comparison of selected portfolios. Users can examine asset allocations to the top mutual funds and view gain ratios that quantify improvements in specific ESG metrics compared to a benchmark portfolio. To further enhance transparency and assist with complex terminology, the researchers integrated LLM-driven features: a chatbot that acts as a question-answering advisor and an on-demand explanation feature triggered by text selection. These tools are built using Retrieval-Augmented Generation (RAG) to provide context-aware, real-time clarifications of financial concepts and optimization processes, directly addressing the difficulty that some users may experience when interpreting advanced metrics like tracking error or specific MCDM techniques.
To evaluate the effectiveness of the platform, the research team conducted a user study involving 19 financial experts from Morningstar, Inc. The participants were provided with a manual and given access to the web-based system before completing a questionnaire. The results indicated generally positive feedback regarding the tool’s usability and usefulness. Specifically, a majority of the subjects expressed high satisfaction with the visualization plots and the portfolio dashboard. Many noted that the interactive features, such as the ability to rotate 3D plots and add points to the dashboard for side-by-side comparison, were particularly beneficial for recognizing the trade-offs between different financial objectives. Participants also recognized the system’s potential to improve the transparency of the decision-making process.
However, the study also revealed certain limitations, with some participants noting that the initial version of the interface was more suited to a prototype than a production environment. A recurring issue raised by the experts was the challenge of understanding technical acronyms and terms, such as ASF, PW, and TOPSIS. This feedback directly informed the implementation of the LLM-driven chatbot and on-demand explanation features in the final version of the prototype. The researchers emphasize that these LLM integrations provide a significant enhancement to transparency by offering accessible, real-time clarifications that help bridge the knowledge gap for users who may have varying levels of expertise in multi-objective optimization.
The researchers also investigated the performance of three MOEA approaches—SMS-EMOA, AGE-MOEA2, and NSGA2—to generate the initial set of 100 non-dominated solutions for each return/risk profile. By measuring the hypervolume of the objective space dominated by each set, they determined that the NSGA2 algorithm produced superior Pareto sets for the 20,000 function evaluations performed. This robust optimization foundation ensures that the portfolios presented to users are diverse and effectively balanced across the various objectives. The researchers note that while the current study provides a solid foundation for the tool’s design, future work will focus on gathering feedback from a broader range of user groups, including end clients, and conducting structured studies to quantitatively assess the tool’s impact on investment alignment and the mitigation of potential agency conflicts.
In conclusion, the platform represents a significant step toward making multi-objective financial portfolio management more transparent and user-friendly. By combining interactive 3D visualizations with LLM-powered explanations, the researchers have created an environment where the complex trade-offs between financial performance and ESG goals can be navigated with greater confidence. The study demonstrates that such tools can empower financial advisors to guide their clients through the selection process, effectively reducing cognitive load and allowing for more informed, strategic decision-making. As the research moves toward potential future applications beyond the financial domain, the underlying principles of the system—interactive exploration, side-by-side comparison, and accessible expert guidance—hold promise for enhancing decision-making in other areas of multi-objective optimization.
Beyond broad portfolio goals, the optimization framework specifically models asset distributions across six distinct asset classes: equity in U.S. markets, equity in global markets, equity in emerging markets, bonds in the U.S. market, bonds in global markets, and cash equivalents. Each investor first selects a baseline profile among conservative, moderate, and aggressive allocations, where aggressive settings target higher returns alongside greater risk and conservative choices prioritize stability. Across the 119 mutual funds examined, expert ratings spanned 15 distinct categories divided into five positive and ten negative environmental, social, and governance dimensions, with particularly low baseline scores recorded in human development, palm oil, pesticides, and healthy ecosystems. The multi-objective formulation constrains candidate solutions by holding tracking error deviations under five percent relative to benchmark distributions, while requiring asset allocation weights to be non-negative and sum to one. In the mathematical objective function, the algorithm minimizes tracking error, aggregate negative metrics, and designated negative concerns such as thermal coal and fossil fuels, while simultaneously maximizing aggregate positive metrics and target priorities like climate action and healthy ecosystems.
To guide users through candidate portfolios without enforcing rigid weights, the web application incorporates three distinct knee-point identification strategies alongside interactive multi-criteria decision-making algorithms. The angle-based knee approach determines candidates by calculating angles between neighbor points and selecting the point exhibiting the widest angular separation. The marginal utility technique tracks objective differences across consecutive solutions, highlighting the portfolio with the greatest total absolute variation, whereas the distance-based method identifies solutions positioned farthest from the extreme boundary line connecting maximum and minimum values. For structured preference weighting, the system embeds the augmented scalarization function and pseudo-weight routines from Pymoo, as well as the Technique for Order of Preference by Similarity to Ideal Solution implemented in MOPO-LSI. Within the visual interface built using HTML, JavaScript, and Plotly, positive dimensions are normalized and inverted so that optimal portfolios converge toward a zero origin marked by a plus symbol. Hovering over any data point turns it orange across both 3D scatter plots and 2D projections, while knee selections appear in green and multi-criteria outcomes are displayed in red.
Detailed evaluation metrics from the 19 surveyed financial professionals, comprising 84.2 percent men and 15.8 percent women, provide quantitative depth regarding system usability and performance. Among these participants, 36.8 percent reported being very familiar with sustainability criteria, while 63.2 percent reported moderate familiarity. When rating general ease of use, 21.1 percent categorized the tools as extremely easy, 42.1 percent rated them very easy, 21.1 percent expressed moderate satisfaction, and 15.8 percent indicated slight satisfaction. Regarding the main 3D and 2D visual plots, 15.8 percent of respondents reported extreme satisfaction and 42.1 percent indicated very high satisfaction, matched by identical ratings of 21.1 percent extreme and 42.1 percent very high satisfaction for the portfolio dashboard. Interactive data entry elements received positive ratings from 68.4 percent of participants, while 15.8 percent expressed dissatisfaction. In operational comparisons, the utility-based knee identification and the augmented scalarization method emerged as the most favored portfolio selection options, while rotating the 3D visual space and clicking points into the dashboard ranked as the preferred interactive features.
The technical deployment of the explanatory architecture relies on FlowiseAI and LangChain, which power a question-answering advisor driven by GPT-3.5-turbo operating over curated background documentation and technical definitions. For real-time clarification, client-side JavaScript captures user text selections across the interface and transmits them through a RESTful API to retrieve concise definitions in a pop-up window. In institutional investment settings such as university endowments, these transparent visual mechanisms and explanatory layers help reconcile agency conflicts between institutional decision-makers and beneficiaries. Formal disclosures and side-by-side metric comparisons allow individual stakeholders to audit recommendations and retain full autonomy over investment choices.
Subject of Research: Transparent multi-objective financial portfolio management using interactive 3D visualizations and LLM-driven explanations.
Article Title: Leveraging interactive visualizations and LLM-driven explanations for transparent multi-objective portfolio management
Article References: Zheng, Y., Zhang, J., Shukla, K. N., O’Leary, M., Wang, D. X., & Xu, J. (2025). Leveraging interactive visualizations and LLM-driven explanations for transparent multi-objective portfolio management. Discover Data, 3(1), Article 50. https://doi.org/10.1007/s44248-025-00065-z
Image Credits: AI Generated
DOI: 10.1007/s44248-025-00065-z
Keywords: portfolio optimization, multi-objective optimization, ESG investing, interactive visualization, large language models, transparent decision making, financial technology, mutual funds, MOEA, NSGA-II, financial dashboard, user study, sustainable finance, investment strategy, data transparency, computational finance
News Source: Denise Maddox. (November 16, 2026). Interactive 3D Visualizations and LLMs Enhance Transparent Multi-Objective Financial Portfolio Management. Scienmag.



