Issue Generation from Consumer Complaints through LLMs

• William Barbosa USP
• Marcos Gôlo USP
• Ricardo Marcacini USP
• Solange Rezende USP

DOI: https://doi.org/10.5753/eniac.2026.30631

Resumo

Issue generation transforms raw textual reports, such as consumer complaint narratives, into descriptions to support triage and service monitoring. Issue generation is inherently an open-world task since new problems continuously emerge. Even though Large Language Models (LLMs) make this generative formulation viable, the transformation of narratives into issues remains underexplored. In this paper, we introduce and empirically study issue generation via opinion mining, and propose five LLM-based strategies that explore different input compositions, including full narratives, keywords, and explicit opinion aspects. We evaluate these approaches on four datasets with real complaints from the Consumer Financial Protection Bureau (CFPB). Our results reveal a consistent trade-off between alignment with the human issues. While statistical analysis shows no single strategy strictly dominates, aspect-guided methods offer significant structural advantages by providing explicit opinion units that support enterprise auditing, monitoring, and interpretability.

Referências

Allam, H., Makubvure, L., Gyamfi, B., Graham, K., and Akinwolere, K. (2025). Text classification how machine learning is revolutionizing text categorization. Information.

Bambauer-Sachse, S. and Stuhldreier, S. M. (2026). Effective complaint management on social networks: the impact of response strategy, type and account on consumer reactions. Journal of Services Marketing, pages 1–21.

Barros, J. M. and De Bona, G. (2021). A deep learning approach for aspect sentiment triplet extraction in portuguese. In Brazilian Conference on Intelligent Systems.

Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J. D., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., et al. (2020). Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901.

Cantini, R., Cosentino, C., Marozzo, F., Talia, D., and Trunfio, P. (2025). Harnessing prompt-based large language models for disaster monitoring and automated reporting from social media feedback. Online Social Networks and Media, 45:100295.

De Caigny, A., Coussement, K., De Bock, K. W., and Lessmann, S. (2020). Incorporating textual information in customer churn prediction models based on a convolutional neural network. International Journal of Forecasting, 36(4):1563–1578.

de Lima, V. M. A., Barbosa, J. R., and Marcacini, R. M. (2025). Issue detection and prioritization based on mobile application reviews. Software Quality Journal, 33(1):6.

Gemma, T. (2025). Gemma 3. Google Deep Mind.

Jorge, G. A. Z., Bezerra, D. A., Xavier, C. C., and Pardo, T. A. S. (2024). Multilingual extractive summarization: Investigating state-of-the-art methods for english and brazilian portuguese. In Brazilian Conference on Intelligent Systems. Springer.

Lee, Y., Kim, J., Kim, D., Kho, Y., Kim, Y., and Kang, P. (2023). Painsight: An extendable opinion mining framework for detecting pain points based on online customer reviews. In Proceedings of the 13th Workshop on Computational Approaches to Subjectivity, Sentiment, & Social Media Analysis, pages 215–227.

Lima, B. and Nogueira, T. (2021). Incorporating text specificity into a convolutional neural network for the classification of review perceived helpfulness. In Brazilian Conference on Intelligent Systems, pages 480–495. Springer.

Liu, B. (2012). Sentiment Analysis and Opinion Mining. Springer Nature.

Liu, Y., Iter, D., Xu, Y., Wang, S., Xu, R., and Zhu, C. (2023). G-eval: Nlg evaluation using gpt-4 with better human alignment. In Proceedings of the 2023 conference on empirical methods in natural language processing, pages 2511–2522.

Meta, T. (2024). The llama 3.1 herd of models. arXiv preprint.

Pontiki, M., Galanis, D., Papageorgiou, H., Androutsopoulos, I., Manandhar, S., Al-Smadi, M., Al-Ayyoub, M., Zhao, Y., Qin, B., De Clercq, O., et al. (2016). Semeval-2016 task 5: Aspect based sentiment analysis. In Int. workshop on semantic evaluation.

Song, R., Li, Y., Tian, M., Wang, H., Giunchiglia, F., and Xu, H. (2025). Causal keyword driven reliable text classification with large language model feedback. Information Processing & Management, 62(2):103964.

Sparck Jones, K. (1972). A statistical interpretation of term specificity and its application in retrieval. Journal of documentation, 28(1):11–21.

Team, Q. (2025). Qwen2.5 technical report.

Trawiński, B., Smźtek, M., Telec, Z., and Lasota, T. (2012). Nonparametric statistical analysis for multiple comparison of machine learning regression algorithms. International Journal of Applied Mathematics and Computer Science, 22(4):867–881.

Wan, M., Safavi, T., Jauhar, S. K., Kim, Y., Counts, S., Neville, J., Suri, S., Shah, C., White, R. W., Yang, L., et al. (2024). Tnt-llm: Text mining at scale with large language models. In Conference on Knowledge Discovery and Data Mining.

Watanangura, P., Vanichrudee, S., Minteer, O., Sringamdee, T., Thanngam, N., and Siriborvornratanakul, T. (2023). A comparative survey of text summarization techniques. SN Computer Science, 5(1):47.

Win, M. N., Ravana, S. D., and Shuib, L. (2022). Sentiment attribution analysis with hierarchical classification and automatic aspect categorization on online user reviews. Malaysian Journal of Computer Science, 35(2):89–110.

Zhang, T., Kishore, V., Wu, F., Weinberger, K. Q., and Artzi, Y. (2019). Bertscore: Evaluating text generation with bert. arXiv preprint arXiv:1904.09675.

Zhang, Y., Xu, W., Liu, Z., Nakagawa, T., and Freitag, M. (2025). Penalizing length: Uncovering systematic bias in quality estimation metrics. arXiv e-prints.