Given the limitations of existing sentiment and topic analysis tools, such as heterogeneous and incoherent workflows, limited semantic understanding capabilities, and lack of sensitivity to dynamic evolution, this paper proposes and implements an integrated natural language processing platform system. First, the system unifies and integrates multi-source heterogeneous data using a configurable preprocessing pipeline. Second, it applies the BERT (Bidirectional Encoder Representations from Transformers) model for context-aware sentiment analysis and employs the BERTopic dynamic topic modeling method, which combines UMAP dimensionality reduction and HDBSCAN semantic clustering to model topic evolution over time. Finally, all modules are integrated into a microservice architecture. Experimental results show that the system achieves an F1 score of 88.5% in sentiment analysis, improves topic consistency score by 51.3%, and runs stably under 200 concurrent tasks. This system enables collaborative, dynamic, and efficient sentiment and topic analysis, providing comprehensive and in-depth information support for decision-making.
Wu et al. (Thu,) studied this question.