Explainable Behavioral Analytics for Financial Crime Detection Using Large Language Models

Financial crime continues to evolve through increasingly sophisticated behavioral patterns that are difficult to detect using conventional rule-based monitoring systems. Advances in large language models (LLMs) have created new opportunities for analyzing unstructured textual information, behavioral indicators, and contextual evidence associated with suspicious financial activities. This paper proposes an Explainable Behavioral Analytics framework that combines large language models with machine learning techniques to improve financial crime detection while ensuring transparency and interpretability. The framework integrates structured transaction data with unstructured textual information, including investigation reports, customer interactions, suspicious activity narratives, and compliance documentation. Behavioral features extracted using transformer-based language models are combined with supervised classification algorithms to identify potentially fraudulent activities. Explainability mechanisms are incorporated to provide investigators with understandable reasoning behind model predictions, thereby improving trust, regulatory compliance, and operational decision-making. Experimental analysis using representative financial crime datasets demonstrates that the proposed framework enhances behavioral pattern recognition while reducing false positives and supporting more efficient investigation workflows. The study highlights the growing role of explainable generative artificial intelligence in financial crime analytics and provides practical guidance for integrating language models into enterprise compliance environments.

Financial Crime Detection, Large Language Models, Explainable AI, Behavioral Analytics, Natural Language Processing, Compliance Analytics, Financial Intelligence.

B. Liang et al., “Enhancing aspect-based sentiment analysis with supervised contrastive learning,” in Proc. 30th ACM Int. Conf. Inf. Knowl. Manage., 2021, pp. 3242–3247.

B. Liang, H. Su, L. Gui, E. Cambria, and R. Xu, “Aspect-based sentiment analysis via affective knowledge enhanced graph convolutional networks,” Knowledge-Based Syst., vol. 235, Jan.2022, Art. no. 107643.

B. Wang, L. Ding, Q. Zhong, X. Li, and D. Tao, “A contrastive cross-channel data augmentation framework for aspect-based sentiment analysis,” in Proc. 29th Int. Conf. Comput. Ling., 2022, pp. 6691–6704.

E. Cambria, X. Zhang, R. Mao, M. Chen, and K. Kwok, “SenticNet 8: Fusing emotion AI and commonsense AI for interpretable, trustworthy, and explainable affective computing,” in Proc. Int. Conf. Hum.-Comput. Interact. (HCII), 2024, pp. 1–20.

J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “BERT: Pre-training of deep bidirectional transformers for language understanding,” in Proc. Conf. North Am. Chapter Assoc. Comput. Ling.: Hum. Lang. Technol., 2019, pp. 4171–4186.

J. Ye et al., “LLM-DA: Data augmentation via large language models for few-shot named entity recognition,” 2024, arXiv:2402.14568.

L. Xu and W. Wang, “Improving aspect-based sentiment analysis with contrastive learning,” Nat. Lang. Process. J., vol. 3, Jun.2023, Art. no. 100009.

M. Pontiki et al., “SemEval-2016 task 5: Aspect based sentiment analysis,” in Proc. ProWorkshop on Semant. Eval. (SemEval), 2016, pp. 19–30.

Q. Cheng, X. Yang, T. Sun, L. Li, and X. Qiu, “Improving contrastive learning of sentence embeddings from AI feedback,” in Proc. Find. Assoc. Comput. Ling. (ACL), 2023, pp. 11122–11138.

S. Dhuliawala et al., “Chain-of-verification reduces hallucination in large language models,” 2023, arXiv:2309.11495.

T. Chen, S. Kornblith, M. Norouzi, and G. Hinton, “A simple framework for contrastive learning of visual representations,” in Proc. Int. Conf. Mach. Learn., 2020, pp. 1597–1607.

Y. Ma, H. Peng, and E. Cambria, “Targeted aspect-based sentiment analysis via embedding commonsense knowledge into an attentive LSTM,” in Proc. 32nd AAAI Conf. Artif. Intel., 2018, pp. 5876–5883.

Y. Ma, H. Peng, T. Khan, E. Cambria, and A. Hussain, “Sentic LSTM: a hybrid network for targeted aspect-based sentiment analysis,” Cogn. Comput., vol. 10, no. 8, 2018, pp. 639–650.

Z. Li, Y. Zou, C. Zhang, Q. Zhang, and Z. Wei, “Learning implicit sentiment in aspect-based sentiment analysis with supervised contrastive pre-training,” in Proc. Conf. Empirical Methods Nat. Lang. Process., 2021, pp. 246–256.