Graph-based method improves document compliance checking accuracy
UCVE: Graph Structured Retrieval for Compliance Validation
Computation and Language
Summary
Checking if a document meets all specific rules can be tricky when rules depend on each other. The authors introduce a new way called UCVE that sees these rules as connected points in a graph. By learning how these connections matter, their method better finds which rules a document meets or breaks. Tested in different areas, this approach consistently improved accuracy over older methods that look at rules one by one.
What this means in practice
- •For legal compliance teams: Enhance document review by accurately identifying all applicable regulatory obligations and violations with graph-structured retrieval.
- •For risk management analysts: Improve detection of unmet requirements in complex compliance documents through learned graph relationships among obligations.
Authors
Emily Chen, Michelle Menachery, Himanshu Kumar, Rahul Nenavath, Amirfarrokh Iranitalab
Abstract
Structured compliance validation requires determining whether every obligation in a predefined requirement space is satisfied, violated, or unaddressed by a given document. This can be described as a coverage problem that existing retrieval and classification systems, by treating requirements as independent, are not designed to solve. We introduce \textbf{\ac{UCVE}}, a domain-agnostic framework that represents compliance requirement spaces as a \ac{CRG} and performs retrieval via spectral reranking over its reinforcement learning-weighted Laplacian. Edges between requirement nodes are learned via reinforcement learning with a traversal reward grounded in violation detection, reshaping the graph's community structure such that low-pass spectral filtering distinguishes relevant from irrelevant obligation clusters. Evaluating across three compliance domains, \ac{UCVE} averages 4.3\% matched-$K$ recall lift over flat cosine retrieval and consistently outperforms strong retrieval and graph-based baselines.