Sentiment analysis and satisfaction evaluation of ecotourism consumption under a multi-source deep learning framework

Ecotourism promotes sustainable development, but evaluating tourists’ experiences becomes difficult because of different multi-source feedback. Traditional methods using text or rating have been inaccurate. Research incorporates textual reviews and numerical ratings to enhance precision, depth and interpretability in ecotourism sentiments and satisfaction evaluation. The primary goal is to develop and validate an intelligent multi-source framework capable of robust sentiment classification and satisfaction prediction. To achieve this, this research proposes a Green Anaconda Optimized Long Short-Term Transformer Memory (GAO-LSTR) model that integrates the established LSTM–Transformer hybrid model with Green Anaconda Optimization (GAO) to improve parameter optimization, contextual feature learning, and multi-source sentiment classification performance. Multi-source data is gathered from ecotourism platforms, online review systems, and social media sources, including the Ecotourism Sentiment and Satisfaction Dataset. Text preprocessing includes tokenization and stop-word removal. TF-IDF is used for textual feature vectorization, while BERT is utilized to generate contextualized semantic embeddings from textual reviews, capturing deep linguistic and contextual relationships for improved feature representation. Long Short-Term Memory (LSTM) layers capture sequential linguistic patterns, and Transformer components model long-range dependencies. The GAO is employed to optimize the hyperparameters of the LSTR, while network weights are learned through Adam-based backpropagation, thereby improving convergence stability, robustness, and predictive performance. The framework is implemented in Python using TensorFlow and PyTorch. On the benchmark Ecotourism Sentiment and Satisfaction Dataset containing structured sentiment annotations, the proposed GAO-LSTR model achieved an accuracy of 98.13% for sentiment classification and 79% for the satisfaction prediction task. The combined usage of textual and numerical characteristics guarantees high robustness, better interpretability, and consistently high prediction quality. The framework significantly enhances ecotourism sentiment and satisfaction evaluation by effectively integrating heterogeneous data modalities. The GAO-LSTR model provides a scalable and reliable solution for real-world ecotourism analytics, supporting improved decision-making and promoting sustainable tourism development.