Feedback-based self-learning in large-scale conversational AI agents
Open accessAbstract
Today, most of the large-scale conversational AI agents such as Alexa, Siri, or Google Assistant are built using manually annotated data to train the different components of the system including automatic speech recognition (ASR), natural language understanding (NLU), and entity resolution (ER). Typically, the accuracy of the machine learning models in these components are improved by manually transcribing and annotating data. As the scope of these systems increase to cover more scenarios and domains, manual annotation to improve the accuracy of these components becomes prohibitively costly and time con-suming. In this paper, we propose a system that leverages customer/system interaction feedback signals to automate learning without any manual annotation. Users of these systems tend to modify a previous query in hopes of fixing an error in the previous turn to get the right results. These reformulations, which are often preceded by defective experiences caused by either errors in ASR, NLU, ER, or the application. In some cases, users may not properly formulate their requests (e.g., providing partial title of a song), but gleaning across a wider pool of users and sessions reveals the underlying recurrent patterns. Our proposed self-learning system automatically detects the errors, generates reformulations, and deploys fixes to the runtime system to correct different types of errors occurring in different components of the system. In particular, we propose leveraging an absorbing Markov Chain model as a collaborative filtering mechanism in a novel attempt to mine these patterns, and coupling it with a guardrail rewrite selection mechanism that reactively evaluates these fixes using feedback friction data. We show that our approach is highly scalable, and able to learn reformulations that reduce Alexa-user errors by pooling anonymized data across millions of customers. The proposed self-learning system achieves a win-loss ratio of 11.8 and effectively reduces the defect rate by more than 30 percent on utterance level reformulations in our production A/B tests. To the best of our knowledge, this is the first self-learning large-scale conversational AI system in production.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Some AI Agent systems integrate basic learning capabilities through feedback loops, heuristics, or updated context buffers to refine behavior over time, particularly in personalized recommendations or conversation flow management.
“Some systems integrate basic learning capabilities [72] through feedback loops [73,74], heuristics [75], or updated context buffers to refine behavior over time, particularly in settings like personalized recommendations or conversation flow management [7678].”
From AI Agents vs. Agentic AI: A Conceptual taxonomy, applications and challenges (2026) · cites this paper as [73]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
Sequence to Sequence Learning with Neural Networks
Sutskever, I., Vinyals, O., & Le, Q. V. (2014) · arXiv · doi:10.48550/arxiv.1409.3215 · Cited by 17,766 · see citing works
Get To The Point: Summarization with Pointer-Generator Networks
See, A., Liu, P. J., & Manning, C. D. (2017) · Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · doi:10.18653/v1/p17-1099 · Cited by 3,775 · see citing works
THOMPSON, W. R. (1933) · Biometrika · doi:10.1093/biomet/25.3-4.285 · Cited by 2,712 · see citing works
A contextual-bandit approach to personalized news article recommendation
Li, L., Chu, W., Langford, J., & Schapire, R. E. (2010) · Proceedings of the 19th international conference on World wide web · doi:10.1145/1772690.1772758 · Cited by 2,562 · see citing works
(2003) · Informa Healthcare eBooks · doi:10.1201/9780203912799.ch3 · Cited by 1,208 · see citing works
(2011) · Cited by 1,000 · see citing works
Weighted finite-state transducers in speech recognition
Mohri, M., Pereira, F. M. B., & Riley, M. (2002) · Computer Speech & Language · doi:10.1006/csla.2001.0184 · Cited by 896 · see citing works
Fahrmeir, L., & Kaufmann, H. (1985) · The Annals of Statistics · doi:10.1214/aos/1176346597 · Cited by 520 · see citing works
(2010) · Cited by 444 · see citing works
(1998) · Taylor & Francis eBooks · doi:10.1201/noe0750305136.ch5 · Cited by 412 · see citing works
Neural Approaches to Conversational AI
Gao, J., Galley, M., & Li, L. (2018) · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · doi:10.1145/3209978.3210183 · Cited by 409 · see citing works
Beyond Recommender Systems: Helping People Help Each Other
Terveen, L., & Hill, W. (2001) · Cited by 248 · see citing works
Sarikaya, R. (2017) · IEEE Signal Processing Magazine · doi:10.1109/msp.2016.2617341 · Cited by 196 · see citing works
Self-citationNeural Approaches to Conversational AIPreprint
Gao, J., Galley, M., & Li, L. C. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1809.08267 · Cited by 135 · see citing works
Neural Approaches to Conversational AI
Gao, J., Galley, M., & Li, L. (2019) · now publishers, Inc. eBooks · doi:10.1561/9781680835533 · Cited by 124 · see citing works
A Hidden Markov Model for Collaborative Filtering1
Sahoo, N., Singh, & Mukhopadhyay, T. (2012) · MIS Quarterly · doi:10.2307/41703509 · Cited by 123 · see citing works
An Efficient Bandit Algorithm for Realtime Multivariate Optimization
Hill, D. N., Nassif, H., Liu, Y., Iyer, A., & Vishwanathan, S. V. N. (2017) · Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/3097983.3098184 · Cited by 112 · see citing works
The Limiting Distributions of Certain Statistics
Doob, J. L. (1935) · The Annals of Mathematical Statistics · doi:10.1214/aoms/1177732594 · Cited by 109 · see citing works
A Hidden Markov Model for Collaborative FilteringPreprint
Sahoo, N., Singh, P. V., & Mukhopadhyay, T. (2010) · SSRN Electronic Journal · doi:10.2139/ssrn.1700585 · Cited by 107 · see citing works
Patterns of query reformulation during Web searching
Jansen, B. J. J., Booth, D. L., & Spink, A. H. (2009) · Journal of the American Society for Information Science and Technology · doi:10.1002/asi.21071 · Cited by 92 · see citing works
A Markov chain collaborative filtering model for course enrollment recommendations
Khorasani, E. S., Zhenge, Z., & Champaign, J. (2016) · IEEE International Conference on Big Data (BigData) · doi:10.1109/bigdata.2016.7841011 · Cited by 30 · see citing works
WEB RECOMMENDATION SYSTEM BASED ON A MARKOV-CHAIN MODEL
Fouss, F., Faulkner, S., Kolp, M., Pirotte, A., & Saerens, M. (2005) · International Conference on Enterprise Information Systems · doi:10.5220/0002550700560063 · Cited by 6 · see citing works
BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2018) · arXiv · doi:10.48550/arxiv.1810.04805
Flagged first, then by citations
1 of the 24 references couldn’t be matched to a known paper.
Research works printed but not linked: 2
- Zaharia, M., R. S. Xin, P. Wendell, T. Das, M. Armbrust, A. Dave, X. Meng, et al. 2016. "Apache spark: A unified engine for big data processing." Communications of the ACM 59(11): 56-65.
- Zhu, X., J. Guo, X. Cheng, and Y. Lan. 2012. "More than rel- evance: High utility query recommendation by mining users' search behaviors." In CIKM'12, October 29-November 2, Maui, HI, USA.
Cited by
Sorted by citations
AI Agents vs. Agentic AI: A Conceptual taxonomy, applications and challenges
Sapkota, R., Roumeliotis, K. I., & Karkee, M. (2026) · Information Fusion · doi:10.1016/j.inffus.2025.103599 · Cited by 244 · see citing works
AI Agents vs. Agentic AI: A Conceptual Taxonomy, Applications and Challenges
Sapkota, R., Roumeliotis, K. I., & Karkee, M. (2025) · SuperIntelligence - Robotics - Safety & Alignment · doi:10.70777/si.v2i3.15161 · Cited by 186 · see citing works
Huang, Z., Li, K., Jiang, Y., Jia, Z., Lv, L., & Ma, Y. (2024) · Knowledge-Based Systems · doi:10.1016/j.knosys.2024.112311 · Cited by 89 · see citing works
Manager Appraisal of Artificial Intelligence Investments
Queiroz, M., Anand, A., & Baird, A. M. (2024) · Journal of Management Information Systems · doi:10.1080/07421222.2024.2376383 · Cited by 28 · see citing works
Park, S., Li, H., Patel, A., Mudgal, S., Lee, S., Kim, Y., Matsoukas, S., & Sarikaya, R. (2021) · Conference on Empirical Methods in Natural Language Processing (EMNLP) · doi:10.18653/v1/2021.emnlp-main.489 · Cited by 18 · see citing works
Self-citationYamamoto, M., Xu, S., Kee, K. F., & Li, W. (2025) · Journal of Trust Research · doi:10.1080/21515581.2024.2445505 · Cited by 10 · see citing works
Hayawi, K., & Shahriar, S. (2024) · Preprints.org · doi:10.20944/preprints202404.0709.v1 · Cited by 9 · see citing works
To learn or not learn from AI? Unpacking the effects of feedback valence on novel insights recall
Turel, O. (2024) · European Journal of Information Systems · doi:10.1080/0960085x.2024.2426473 · Cited by 8 · see citing works
Pre-Training for Query Rewriting in A Spoken Language Understanding SystemPreprint
Chen, Z., Fan, X., Yuan, L., Mathias, L., & Guo, C. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2002.05607 · Cited by 4 · see citing works
Self-citationAn Epistemology on Role of AI-Powered E-commerce Chatbot to Support CX
Suryawanshi, S. K., & Gohil, J. A. (2023) · Algorithms for intelligent systems · doi:10.1007/978-981-99-3485-0_9 · Cited by 3 · see citing works
Integrating Maintenance Practices with Industry 4.0 in Manufacturing System
Kufre, E. U., Onokwai, A. O., & Onifade, M. K. (2024) · International Conference on Science, Engineering and Business for Driving Sustainable Development Goals (SEB4SDG) · doi:10.1109/seb4sdg60871.2024.10630010 · Cited by 3 · see citing works
Review on Machine Learning-Based Mock Interview Models
Shirodkar, M., Parab, V., Chavan, C., Shende, J., & Sawant, V. (2024) · International Conference on Computing, Communication and Networking Technologies (ICCCNT) · doi:10.1109/icccnt61001.2024.10724271 · Cited by 3 · see citing works
Ge, Z.-G. (2026) · Innovations in Education and Teaching International · doi:10.1080/14703297.2026.2620543 · Cited by 3 · see citing works
When noise mitigates bias in human–algorithm decision-making: An agent-based model
Parsons, S. P., & Torenvlied, R. (2025) · PLoS ONE · doi:10.1371/journal.pone.0339273 · Cited by 1 · see citing works
Zia, M. F., Ouameur, M. A., Bagaa, M., Massicotte, D., & Ksentini, A. (2026) · Computer Networks · doi:10.1016/j.comnet.2026.112142 · Cited by 2 · see citing works
Text Generation with GAN Networks using Feedback Score
Kuznetsov, D. (2023) · International Journal on Cybernetics & Informatics · doi:10.5121/ijci.2023.120301
Integrating Maintenance 4.0 in Manufacturing System: A Multi-Criteria Approach
Kufre, E. U., Ajayi, S. T., Olanrewaju, O. A., Ighravwe, D. E., Amole, A. O., & Etangetuk, E. A. (2024) · IEEE Nigeria International Conference on Disruptive Technologies for Sustainable Development (NIGERCON) · doi:10.1109/nigercon62786.2024.10927220
Survey of Agents Methodologies in the Financial Domain
Luqman, M., Gauba, H., Kumar, P., Desai, A. P., Yadav, A., Prajapati, R., & Yadav, P. (2025) · IEEE International Conference on Big Data (BigData) · doi:10.1109/bigdata66926.2025.11402262
Yang, Y., Wang, T., Pan, M., Zhang, Y., Yue, Y., & Fu, M. (2026) · Science China Information Sciences · doi:10.1007/s11432-025-4852-4
Operationalizing Trustworthy AI Agents: Reliability, Safety and Governance in Real-World Deployments
Kota, S. K. (2026) · International Conference on Inventive Computing and Informatics (ICICI) · doi:10.1109/icici68773.2026.11580693
Tefo, R. M., & Goosen, L. (2026) · Advances in computational intelligence and robotics book series · doi:10.4018/979-8-3373-7225-9.ch005
- Authors
- Ponnusamy, P., Ghias, A., Yi, Y., Yao, B., Guo, C., & Sarikaya, R.
- Affiliations
- Amazon (United States) (US)
- Venue
- AI Magazine
- Publisher
- Wiley
- Published
- 2022-01-12
- Volume
- 42
- Issue
- 4
- Pages
- 43-56
- ISSN
- 0738-4602 (ISSN-L)
- 2371-9621
- OpenAlex
- W2987278351
- Cited by
- 19
- References
- 24
- Citations by year
- 2026: 5 · 2025: 3 · 2024: 7 · 2023: 2 · 2021: 1 · 2020: 1
- FWCI
- 0.99
- Type
- journal article
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- AI Magazine (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Speech and dialogue systems
- Physical Sciences › Computer Science › Artificial Intelligence › AI in Service Interactions
- Physical Sciences › Computer Science › Artificial Intelligence › Natural Language Processing Techniques
- Keywords
- conversational AI agents, self-learning, user feedback, query rewriting, speech recognition errors, absorbing Markov chains, collaborative filtering, defect detection, Amazon Alexa, annotation-free learning
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
