Aligning Language Models with Investor and Market Behavior for Financial Recommendations

TL;DR

FLARKO addresses the gap where profit-focused financial advice fails to account for user preferences and regulatory constraints. It unifies LLM reasoning with two structured knowledge graphs—PKG for investor behavior and MKG for market signals—and optimizes decisions using Kahneman-Tversky Optimization to produce behaviorally aligned, profitable asset recommendations. The framework supports both centralized (CenFLARKO) and federated (FedFLARKO) deployments and demonstrates that mid-sized LLMs can achieve strong performance efficiently, especially when guided by PKG and MKG inputs. Empirical results on the FAR-Trans dataset show improvements in behavioral alignment and joint profitability, with robust performance under non-IID federated settings, underscoring practical potential for privacy-preserving, personalized financial decision-support systems.

Abstract

Most financial recommendation systems often fail to account for key behavioral and regulatory factors, leading to advice that is misaligned with user preferences, difficult to interpret, or unlikely to be followed. We present FLARKO (Financial Language-model for Asset Recommendation with Knowledge-graph Optimization), a novel framework that integrates Large Language Models (LLMs), Knowledge Graphs (KGs), and Kahneman-Tversky Optimization (KTO) to generate asset recommendations that are both profitable and behaviorally aligned. FLARKO encodes users' transaction histories and asset trends as structured KGs, providing interpretable and controllable context for the LLM. To demonstrate the adaptability of our approach, we develop and evaluate both a centralized architecture (CenFLARKO) and a federated variant (FedFLARKO). To our knowledge, this is the first demonstration of combining KTO for fine-tuning of LLMs for financial asset recommendation. We also present the first use of structured KGs to ground LLM reasoning over behavioral financial data in a federated learning (FL) setting. Evaluated on the FAR-Trans dataset, FLARKO consistently outperforms state-of-the-art recommendation baselines on behavioral alignment and joint profitability, while remaining interpretable and resource-efficient.

Figures (3)

• Figure 1: Performance Comparison of CenFLARKO and FedFLARKO Against Baseline Models The left panel plots preference alignment (Pref@3) against profitability (Prof@3) for all models. Prof@3 captures the ability to recommend assets that generate positive returns over the next 180 days, while Pref@3 reflects behavioral alignment by checking whether the user actually purchased the recommended assets. Although some models (e.g., LightGBM) achieve higher profitability, FLARKO-based models, particularly CenFLARKO, excel in preference alignment, demonstrating stronger user-centric performance. The right panel reports Comb@3, which quantifies how often the model recommends assets that are both profitable and behaviorally aligned, representing actionable, high-quality financial advice. Both CenFLARKO and FedFLARKO outperform all baselines on this crucial metric, validating the strength of our approach in real-world financial asset recommendation scenarios.
• Figure 2: Example user transaction in the PKG.
• Figure 3: An example asset summary in the MKG.