Comparative Analysis of Implicit Sentiment Detection with Causal and Prompt-Based LLMs

• Marco Antônio Martins Ribeiro de Jesus UFLA
• Ahmed Esmin UFLA

DOI: https://doi.org/10.5753/eniac.2025.14104

Abstract

Implicit Sentiment Analysis (ISA) remains a challenging NLP problem, as models frequently rely on superficial shortcuts rather than deep contextual cues. This paper directly contrasts two paradigms: a specialized causal model named CLEAN, designed for robustness against spurious correlations and built on a BERT backbone, and a suite of modern open-source large language models (LLMs) such as Llama-3, Gemma-3, Qwen-3, and DeepSeek-R1, executed locally via a streamlined deployment framework. Experiments using widely recognized benchmarks for sentiment analysis reveal that, although prompted LLMs markedly outperform traditional fine-tuning, the causal CLEAN model retains a robustness advantage on the most subtle implicit cases. Our analysis clarifies current trade-offs between the broad versatility of LLMs and the targeted precision of causal methods. As future work, we highlight three directions: (i) combining causal regularization techniques with parameter-efficient fine-tuning approaches like low-rank adaptation methods to fuse both strengths, (ii) extending evaluation to cross-domain and multilingual ISA scenarios, and (iii) integrating explanation-based feedback loops to further reduce shortcut learning observed in prior approaches to sentiment analysis.

References

Devlin, J.; Chang, M. W.; Lee, K.; Toutanova, K. (2019). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. arXiv:1810.04805.

Gemma Team; Mesnard, T.; Hardin, C.; et al. (2024). Gemma: Open Models Based on Gemini Research and Technology. arXiv:2403.08295.

Guo, D.; Yang, D.; Zhang, H.; et al. (2025). DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning. arXiv:2501.12948.

Llama Team. (2024). The Llama 3 Herd of Models. Meta AI. Available at: [link]

Liu, J.; Peng, B.; Shao, Z.; Wang, X.; Wang, Y. (2024). Ollama: Large Language Models Made Easy. arXiv:2405.02257.

Pontiki, M.; Galanis, D.; Pavlopoulos, J.; et al. (2014). SemEval-2014 Task 4: Aspect-Based Sentiment Analysis. In Proc. SemEval 2014, pp. 27–35.

Qwen Team; Yang, A.; Li, A.; et al. (2025). Qwen3 Technical Report. arXiv:2505.09388.

Radford, A.; Wu, J.; Child, R.; et al. (2019). Language Models are Unsupervised Multitask Learners. OpenAI Blog 1(8).

Russo, I.; Caselli, T.; Strapparava, C. (2015). SemEval-2015 Task 9: CLIPEval Implicit Polarity of Events. In Proc. SemEval 2015, pp. 450–454.

Wang, S.; Zhou, J.; Sun, C.; et al. (2022). Causal Intervention Improves Implicit Sentiment Analysis. In Proc. COLING 2022, pp. 6966–6977.

Esuli, A.; Sebastiani, F. (2006). SENTIWORDNET: A Publicly Available Lexical Resource for Opinion Mining. In Proc. LREC 2006.

Speer, R.; Chin, J.; Havasi, C. (2017). ConceptNet 5.5: An Open Multilingual Graph of General Knowledge. In Proc. AAAI-17.

Mao, Y.; Liu, Q.; Zhang, Y. (2025). Enhancing Implicit Sentiment Analysis via Knowledge Enhancement and Context Information. Complex & Intelligent Systems, 11, Article 222.

Yang, S.; Xing, L.; Li, Y.; Chang, Z. (2021). Implicit Sentiment Analysis Based on Graph Attention Neural Network. Engineering Reports, 3:e12452.

Zhao, Y.; Mamat, M.; Aysa, A.; Ubul, K. (2024). A Dynamic Graph Structural Framework for Implicit Sentiment Identification Based on Complementary Semantic and Structural Information. Scientific Reports, 14, 16563.

Riloff, E.; Qadir, A.; Surve, P.; et al. (2013). Sarcasm as Contrast Between a Positive Sentiment and Negative Situation. In Proc. EMNLP 2013, pp. 704–714.

Filatova, E. (2017). Sarcasm Detection Using Sentiment Flow Shifts. In Proc. FLAIRS 30, Florida, USA.

Houlsby, N.; Giurgiu, A.; Jastrzebski, S.; et al. (2019). Parameter-Efficient Transfer Learning for NLP. In Proceedings of the 36th International Conference on Machine Learning (ICML 2019), pp. 2790–2799.

Hu, E.; Shen, Y.; Wallis, P.; Allen-Zhu, Z.; Li, Y.; Wang, L.; Chen, W. (2022). LoRA: Low-Rank Adaptation of Large Language Models. arXiv:2106.09685.

Hernán, Miguel A.; Robins, James M. (2020). Causal Inference: What If. Chapman & Hall/CRC, Boca Raton, FL. Available at [link].

Pearl, Judea. (2009). Causality: Models, Reasoning and Inference (2nd ed.). Cambridge University Press, New York.

Fan, F.; Feng, Y.; Zhao, D. (2018). Multi-grained Attention Network for Aspect-Level Sentiment Classification. In Proceedings of EMNLP 2018, pp. 3433–3442, Brussels, Belgium.

Sun, Y.; Li, J.; Wang, L.; Liu, X. (2019). Convolution over Dependency Tree for Aspect-Level Sentiment Classification. In Proceedings of ACL 2019, pp. 2304–2314, Florence, Italy.

Fei, H.; Li, B.; Liu, Q.; Bing, L.; Chua, T. S. (2023). Reasoning Implicit Sentiment with Chain-of-Thought Prompting. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics – Volume 2 (Short Papers), 1171–1182.

Zhou, X.; Obeid, O.; Ng, M. K. (2023). Implicit Counterfactual Data Augmentation for Robust Learning. arXiv preprint arXiv:2304.13431.

Oprea, S. V.; Magdy, W. (2020). iSarcasm: A Dataset of Intended Sarcasm. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 1279–1289.

Xiao, Z. C.; Wu, J. J.; Chen, Q. C.; Deng, C. K. (2021). BERT4GCN: Using BERT Intermediate Layers to Augment GCN for Aspect-Based Sentiment Classification. arXiv preprint arXiv:2110.00171.

Zhang, W.; Li, X.; Bing, L.; Lam, W. (2021). Cross-Lingual Aspect-Based Sentiment Analysis with Multilingual Language Models. Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, 9206–9218.