Abstract:Soundscape perception is an important entry point for understanding the public'senvironmental experience and improving the quality of national park management. Withthe development of social media, online comment texts have become a new type of datasource for obtaining soundscape perception information. Taking the first four national parks in China as the research object, this research used text mining (Jieba split-word technique),sentiment analysis model (LSTM+Word2vec), and time series analysis model (ARIMA) to construct a framework for quantifying and predicting soundscape perception based ononline texts, and to explore the public's sentiment tendency towards the soundscape ofthe national parks and its dynamic change. The study found that the public prefers naturalsound, of which geophysical sound accounts for the highest percentage, artificial sound accounts for a higher proportion but presents ambivalence, biological sound accounts fora lower proportion but has a positive impact on the ecological experience. The difference in emotional perception of different national parks is obvious, with Wuyishan National Parkhaving the highest percentage of positive emotions (73.18%), and Hainan Tropical Rainforest Park showing a clear tendency to negative emotion influenced by noise; The ARIMA modelprediction results show that, except for the Hainan Tropical Rainforest National Park, the future emotional tendency of other parks is better overall, Wuyishan National Park has themost stable emotional tendency, the reliability of the prediction results of the Giant PandaNational Park is higher, and the confidence interval of the prediction of the Sanjiangyuan National Park is larger. Based on the research results, three recommendations were madeto add soundscape corridors and characteristic soundscape experiences, strengthen noise management, and establish a smart soundscape management system.
