Application of Sentiment Analysis of user Comments Combined with Natural Language Processing in the Evaluation of New Media Operation Effect

This study integrates RoBERTa (Robustly Optimized BERT Pretraining Approach) sentiment analysis with LDA (Latent Dirichlet Allocation) topic mining to evaluate new media operations. User comments were crawled, and RoBERTa was fine-tuned for sentiment scoring adapted to new media language. LDA extracted topics, analyzing sentiment distributions to construct a matrix. ARIMA (Auto Regressive Integrated Moving Average) tracked sentiment trends, revealing dynamic user needs. Results showed 60% positive sentiment and 75% satisfaction coverage under the current strategy, with room for improvement. This method identifies explicit emotions and implicit demands through topic-driven sentiment dynamics, offering insights for strategy optimization. By synergizing advanced NLP models, it provides a granular understanding of user feedback, enabling data-driven decisions to enhance user-centric engagement. This approach bridges surface metrics and deeper user insights, improving operational effectiveness.