Publication Date

Spring 2026

Degree Type

Master's Project

Degree Name

Master of Science in Computer Science (MSCS)

Department

Computer Science

First Advisor

Katerina Potika

Second Advisor

Robert Chun

Third Advisor

Prithvi Pratahkal

Keywords

Temporal Financial Knowledge Graphs, Financial Insights, Link Prediction, Neuro Symbolic Reasoning

Abstract

Financial markets react rapidly to events, making the timely interpretation of unstructured financial text critical for analysts and automated systems. Recently, Temporal Knowledge Graphs (TKGs) have been used to store financial information in a structured manner, enabling querying and predicting future interactions. We focus on predicting links (relations) at a specific time using TKGs, which will help us gain financial insight into future steps. In this work, we investigate how combining graph neural networks and symbolic reasoning approaches over Temporal Knowledge Graphs improves reasoning capabili- ties and predicts future quadruples (triples and time) in TKGs. We propose Financial KG Reasoner (FinKGR), a transformer-based model that uses a neuro-symbolic ap- proach to reason over TKGs. While neural reasoning learns to rank entities based on KG structural similarity and temporal patterns, symbolic reasoning mines multi-hop relational rules from the training graph and applies them via forward chaining to produce a fixed set of rule-derived inferred triples. Combining both methods improves coverage in the financial domain, because each component models a different aspect of how financial facts spread through connected relationships. The neuro-symbolic union achieves higher coverage than the neural model, demonstrating that the two components are highly complementary.

Download

Available for download on Saturday, May 22, 2027

Share

COinS