Advancements in Natural Language Processing (NLP) technology have progressed rapidly, marked by the emergence of various Large Language Models (LLMs) such as ChatGPT, Gemini, and DeepSeek AI. One particularly popular model is DeepSeek AI due to its ability to understand and respond to natural language text more contextually. The increasing popularity of this application is accompanied by a growing number of user reviews, which serve as an important source of data for capturing their experiences and perceptions. This study aims to analyze user sentiment toward the DeepSeek AI application using a deep learning approach. Specifically, the research focuses on evaluating the performance of sentiment classification models in the context of Indonesian-language data, which is relatively limited and imbalanced. The dataset was collected from user reviews on the Google Play Store and categorized into three sentiment classes: positive, negative, and neutral. The method employed is a combination of IndoBERT and Bidirectional Long Short-Term Memory (BiLSTM). IndoBERT is used to generate contextual text representations in Indonesian, while BiLSTM is utilized to recognize sequential word patterns. Experimental results show that this hybrid model achieves an accuracy of 45%, with the highest F1-score of 0.66 in the positive class. Meanwhile, a macro-average F1-score of 0.33 and a ROC-AUC of 0.546 indicate that the model’s performance remains limited in distinguishing the three classes evenly. Nevertheless, the main contribution of this study lies in the development of a new dataset consisting of 1,774 Indonesian-language reviews related to LLM-based applications, which can be used for further research in the field of Natural Language Processing (NLP). The study also demonstrates the effectiveness of integrating IndoBERT and BiLSTM for sentiment analysis of Indonesian text with imbalanced data distribution.