Detecting Forged Receipts with Domain-Specific Ontology-Based Entities & Relations

Beatriz Martínez Tornés, Emanuela Boros, Antoine Doucet, Petra Gomez-Krämer, Jean-Marc Ogier

Abstract: In this paper, we tackle the task of document fraud detection. We consider that this task can be addressed with natural language processing techniques. We treat it as a regression-based approach, by taking advantage of a pre-trained language model in order to represent the textual content, and by enriching the representation with domain-specific ontology-based entities and relations. We emulate an entity-based approach by comparing different types of input: raw text, extracted entities and a triple-based reformulation of the document content. For our experimental setup, we utilize the single freely available dataset of forged receipts, and we provide a deep analysis of our results in regard to the efficiency of our methods. Our findings show interesting correlations between the types of ontology relations (e.g., has_address, amounts_to), types of entities (product, company, etc.) and the performance of a regression-based language model that could help to study the transfer learning from natural language processing (NLP) methods to boost the performance of existing fraud detection systems.