Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Parallel and Multi-Stage Knowledge Graph Retrieval for Behaviorally Aligned Financial Asset Recommendations

Fernando Spadea, Oshani Seneviratne

TL;DR

This work addresses the challenge of providing trustworthy, behaviorally grounded financial asset recommendations under LLM context limits. It introduces RAG-FLARKO, a two-stage retrieval framework that builds compact, temporally filtered subgraphs from a Personal Transaction KG ($PKG$) and a Market KG ($MKG$) to ground LLM-generated recommendations, reducing token overhead. By leveraging SPARQL-based retrieval and JSON-LD subgraphs, the approach enables smaller models to achieve competitive profitability and behavioral alignment, with strong gains especially for resource-constrained deployments. The results on the FAR-Trans dataset demonstrate that multi-stage, context-aware retrieval improves Comb@3 scores and makes grounded financial AI more practical for edge and federated settings, while offering directions for future integration with symbolic reasoning and richer user personas.

Abstract

Large language models (LLMs) show promise for personalized financial recommendations but are hampered by context limits, hallucinations, and a lack of behavioral grounding. Our prior work, FLARKO, embedded structured knowledge graphs (KGs) in LLM prompts to align advice with user behavior and market data. This paper introduces RAG-FLARKO, a retrieval-augmented extension to FLARKO, that overcomes scalability and relevance challenges using multi-stage and parallel KG retrieval processes. Our method first retrieves behaviorally relevant entities from a user's transaction KG and then uses this context to filter temporally consistent signals from a market KG, constructing a compact, grounded subgraph for the LLM. This pipeline reduces context overhead and sharpens the model's focus on relevant information. Empirical evaluation on a real-world financial transaction dataset demonstrates that RAG-FLARKO significantly enhances recommendation quality. Notably, our framework enables smaller, more efficient models to achieve high performance in both profitability and behavioral alignment, presenting a viable path for deploying grounded financial AI in resource-constrained environments.

Parallel and Multi-Stage Knowledge Graph Retrieval for Behaviorally Aligned Financial Asset Recommendations

TL;DR

This work addresses the challenge of providing trustworthy, behaviorally grounded financial asset recommendations under LLM context limits. It introduces RAG-FLARKO, a two-stage retrieval framework that builds compact, temporally filtered subgraphs from a Personal Transaction KG () and a Market KG () to ground LLM-generated recommendations, reducing token overhead. By leveraging SPARQL-based retrieval and JSON-LD subgraphs, the approach enables smaller models to achieve competitive profitability and behavioral alignment, with strong gains especially for resource-constrained deployments. The results on the FAR-Trans dataset demonstrate that multi-stage, context-aware retrieval improves Comb@3 scores and makes grounded financial AI more practical for edge and federated settings, while offering directions for future integration with symbolic reasoning and richer user personas.

Abstract

Large language models (LLMs) show promise for personalized financial recommendations but are hampered by context limits, hallucinations, and a lack of behavioral grounding. Our prior work, FLARKO, embedded structured knowledge graphs (KGs) in LLM prompts to align advice with user behavior and market data. This paper introduces RAG-FLARKO, a retrieval-augmented extension to FLARKO, that overcomes scalability and relevance challenges using multi-stage and parallel KG retrieval processes. Our method first retrieves behaviorally relevant entities from a user's transaction KG and then uses this context to filter temporally consistent signals from a market KG, constructing a compact, grounded subgraph for the LLM. This pipeline reduces context overhead and sharpens the model's focus on relevant information. Empirical evaluation on a real-world financial transaction dataset demonstrates that RAG-FLARKO significantly enhances recommendation quality. Notably, our framework enables smaller, more efficient models to achieve high performance in both profitability and behavioral alignment, presenting a viable path for deploying grounded financial AI in resource-constrained environments.

Paper Structure

This paper contains 21 sections, 3 figures.

Table of Contents

  1. Introduction
  2. Related Work
  3. KGs in Financial Recommendation Systems
  4. RAG for Financial Tasks
  5. Methodology
  6. Knowledge Graph Design
  7. Multi-Step RAG Pipeline
  8. Dataset
  9. Evaluation
  10. Evaluation Metrics
  11. Baselines and Model Variants
  12. Results
  13. Overall Performance Gains:
  14. Impact of Inter-Stage Context:
  15. Impact of Model Size:
  16. ...and 6 more sections

Figures (3)

  • Figure 1: RAG-FLARKO Multi-Stage Retrieval Pipeline
  • Figure 2: Examples of retrieved subgraphs from the (a) Personal transaction KG and (b) Market KG, each centered on a selected entity.
  • Figure 3: Performance Comparison of FLARKO, RAG-FLARKO, and Parallel RAG-FLARKO The left panel plots preference alignment (Pref@3) against profitability (Prof@3) for all models. The right panel reports Comb@3, which quantifies how often the model recommends assets that are both profitable and behaviorally aligned. The Standard Error of Proportions are plotted in the graphs as well.