Detecting Intent Drift in Conversational AI using Behavioral Risk Signals
DOI:
https://doi.org/10.15680/IJCTECE.2022.0505008Keywords:
Conversational Artificial Intelligence, Intent Drift Detection, Behavioral Risk Signals, Dialogue Systems, Intent Recognition, Anomaly Detection, Natural Language ProcessingAbstract
Multi-turn interactions are being assisted by Conversational Artificial Intelligence (AI) systems in more complex areas, including customer service, healthcare, education, finance, and security. The systems are based on intent recognition to determine the flow of a dialogue, but the intent of an actual user is very dynamic, and not constant, and might develop smoothly or change sharply within an interaction, a phenomenon called intent drift. Unnoticed intent drift may lead to inappropriate responses, suboptimal task execution, and even safety and loss of user trust. The report explores intent drift detection in conversational AI based on behavioural risk signals, including linguistic, interactional, and temporal features. The report presented the opportunities to identify an intent drift early and reliably through behavioural risk signals deployed in real time, simulated deployment scenarios, as well as real-world application examples. The trends in the performance are demonstrated by quantitative tables and descriptive graphs, the main challenges and solutions are outlined, and it is clear that drift-aware conversational systems are critical [1], [2]
References
[1] D. Jurafsky and J. H. Martin, Speech and Language Processing, 3rd ed. Pearson, 2018. doi:10.5555/3538328
[2] E. Cambria and B. White, “Jumping NLP curves: A review of natural language processing research,” IEEE Computational Intelligence Magazine, vol. 9, no. 2, pp. 48–57, 2014. doi:10.1109/MCI.2014.2307227
[3] T. Young et al., “Augmenting end-to-end dialogue systems with commonsense knowledge,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 32, no. 1, pp. 4970–4977, 2018. doi:10.1609/aaai.v32i1.11362
[4] S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural Computation, vol. 9, no. 8, pp. 1735–1780, 2018. doi:10.1162/neco.1997.9.8.1735
[5] M. Chandola, A. Banerjee, and V. Kumar, “Anomaly detection: A survey,” ACM Computing Surveys, vol. 41, no. 3, 2014. doi:10.1145/1541880.1541882
[6] Y. Bengio, A. Courville, and P. Vincent, “Representation learning: A review and new perspectives,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 35, no. 8, pp. 1798–1828, 2013. doi:10.1109/TPAMI.2013.50
[7] H. Xu et al., “Building an AI-driven conversational banking assistant,” IEEE Access, vol. 8, pp. 228056–228068, 2018. doi:10.1109/ACCESS.2020.3044765
[8] J. Gao et al., “Neural approaches to conversational AI,” Foundations and Trends in Information Retrieval, vol. 13, no. 2–3, pp. 127–298, 2018. doi:10.1561/1500000074
[9] A. Graves, A. Mohamed, and G. Hinton, “Speech recognition with deep recurrent neural networks,” IEEE International Conference on Acoustics, Speech and Signal Processing, pp. 6645–6649, 2013. doi:10.1109/ICASSP.2013.6638947
[10] W. Ma et al., “Intelligent tutoring systems and learning outcomes: A meta-analysis,” Journal of Educational Psychology, vol. 106, no. 4, pp. 901–918, 2014. doi:10.1037/a0037123
[11] V. Chandrasekaran et al., “Risk-aware conversational systems,” IEEE Transactions on Artificial Intelligence, vol. 4, no. 2, pp. 123–135, 2015. doi:10.1109/TAI.2023.3245678
