Ensuring that large language models (LLMs) generate responses that resonate with users is critical for successful interactions. This study leverages machine learning techniques to improve chatbot interactions by aligning responses more closely with human preferences. By analyzing the Chatbot Arena dialogue dataset, we identified key features influencing user satisfaction and employed an ensemble model combining Decision Tree, TF-IDF, and BERTopic to enhance prediction accuracy. Our results demonstrate significant improvements in both accuracy and user satisfaction metrics compared to traditional models. This approach addresses the complexities of human-computer dialogue, providing a robust framework for future enhancements. The integration of advanced feature engineering and topic modeling techniques enables the chatbot to generate more contextually relevant and engaging responses, setting a new standard in the field of conversational AI.

In the dynamic landscape of online shopping, predicting user behaviors such as clicks, cart additions, and orders is crucial for optimizing sales strategies. Traditional recommendation systems often focus on a single objective, limiting their effectiveness in a multifaceted e-commerce environment. This article proposes a multi-objective recommendation system that leverages previous events in user sessions to forecast these key metrics, addressing challenges such as imbalanced positive and negative samples, varying session lengths, and the need for effective sampling techniques. Our approach integrates LightGBM, XgBoost, and Recbole GRU4Rec through ensemble learning, combining the strengths of these models to enhance prediction performance. Extensive evaluations demonstrate that our model outperforms existing methods, offering significant improvements in accuracy and robustness. This work provides a comprehensive solution for online retailers to better predict user behaviors and optimize their sales strategies, ultimately enhancing customer satisfaction and business outcomes.

In the rapidly evolving landscape of personalized recommendation systems, accurately predicting user purchase behavior remains a critical challenge. This study presents a novel hybrid model combining LightGBM, DeepFM, and Deep Interest Network (DIN) architectures to enhance the prediction accuracy for the Elo payment dataset. Our approach leverages advanced feature engineering, including clustering and temporal feature extraction, coupled with robust training strategies such as adversarial weight perturbation (AWP) and pseudo-labeling. By integrating these techniques, we achieve superior performance compared to existing models, as measured by AUC (Area Under the Curve) and NDCG (Normalized Discounted Cumulative Gain) metrics. Our model demonstrates a significant improvement in prediction precision, particularly for high-ranking predictions, thereby offering a comprehensive solution to the complexities of user behavior modeling in payment datasets.

Yuyan Chen, Jiyuan Jia, Jiaxin Lu, Siyue Li, Yu Guan, Ming Yang, Qingpei Guo. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

The prevalence of hallucinations in responses generated by large language models (LLMs) poses significant challenges for the reliability of natural language processing applications. This study addresses the detection of such hallucinations through an enhanced Roberta-base model, specifically targeting hallucination responses produced by the Mistral 7B Instruct model. By implementing Low-Rank Adaptation (LoRA) for fine-tuning and incorporating hierarchical multi-head attention and multi-level self-attention weighting mechanisms, we aim to improve both the accuracy of hallucination detection and the interpretability of the model's decisions. Our experimental results demonstrate that the proposed model significantly outperforms baseline models across various metrics, including accuracy, precision, recall, and area under the curve (AUC). Future research directions will explore the integration of larger-scale models and additional fine-tuning techniques to further bolster the model’s capacity for detecting hallucinations, thereby enhancing the reliability of LLM outputs.