Enhancing Delivery Rider Experience: A Knowledge-Enhanced Framework for Short-Text Sentiment Analysis and Topic Mining Using ERNIE 3.0

The rapid expansion of the food delivery industry has highlighted critical challenges in managing delivery riders” satisfaction and operational efficiency. Traditional survey methods fail to capture dynamic sentiment feedback, while semantic sparsity in short-text reviews limits the accuracy of topic modeling. To address these issues, this study proposes a knowledge-enhanced framework combining ERNIE 3.0 and BERTopic for analyzing Chinese short-text reviews from delivery riders. By scraping 5,013 rider reviews from social platforms (e.g., Meituan, Weibo), we achieved high-precision sentiment classification (accuracy: 87.5%, F1 score: 0.85) using ERNIE 3.0, which outperformed BERT by 7% in recognizing ambiguous neutral sentiments through dynamic weighted loss functions. BERTopic extracted five key topics (e.g., salary disputes, order assignment inefficiency), revealing that over 65% of negative sentiments correlate with compensation and workload issues. Temporal analysis further demonstrated that topic evolution (e.g., a 142% surge in salary disputes in Q3 2019) aligns with policy changes and external events like the COVID-19 pandemic. These findings provide actionable insights for platform optimization, validate the effectiveness of knowledge-enhanced models in short-text analysis, and establish a scalable framework for real-time feedback tracking in gig economy management.