The Identification and Analysis of the Centers of Geographical Public Opinions in Flood Disasters Based on Improved Naïve Bayes Network

Abstract

The increasing frequency of floods and the lack of protective measures have the potential to cause severe damage. Working from the perspective of network public opinion is an effective way to understand flood disasters. However, the existing research tends to focus on a single perspective, such as the characteristics of the text, algorithm optimization, or spatial location recognition, while scholars have paid much less attention to the impact of social-psychological differences in space on network public opinion. This research is based on the following hypothesis: When public opinions break out, the differences of network public opinions in geography will form spatially different centers of geographical public opinions in flood disasters (CGeoPOFDs). These centers represent the cities that receive the most attention from network public opinion. Based on this hypothesis, this study proposes a new way of identifying and analyzing CGeoPOFDs. First, two optimization strategies were applied to enhance a naïve Bayes network: syntactic parsing, which was used to optimize the selection of feature word vectors, and ensemble learning, which enabled multi-classifier fusion optimization. Social media data were classified through the improved algorithm, and then, various methods (hotspot analysis, geographic mapping, and sentiment analysis) were used to identify CGeoPOFDs. Finally, analysis was performed in terms of spatiotemporal, virtual, and real dimensions. In addition, microblog social data and real disaster data were used to arrive at empirical results. According to the study findings, the identified CGeoPOFDs offered traditional characteristics of network public opinion while also featuring unique spatiotemporal characteristics. Over time, CGeoPOFDs demonstrated spatial aggregation and bias diffusion and an overall positive emotional tendency.

Modeling Flood Susceptible Areas Using Deep Learning Techniques with Random Subspace: A Case Study of the Mae Chan Basin in Thailand

“…Additionally, they can be integrated with traditional models, which helps overcome their limitations [19]. Some examples of machine learning models are Support Vector Machines (SVM) [20], [21], Artificial Neural Networks (ANN) [22], Classification and Regression Tree (CART) [23], Multilayer Perceptron (MLP) [18], [24], Decision Tree (DT) [20], [25], Random Forest (RF), and Naïve Bayes (NB) [26], [27]. Ensemble and hybrid models, such as ensemble of bagging and Logistic Model Tree (LMT) [28], ensemble of Dagging and Classifier-M5P [29], Reduced Error Pruning Trees with Bagging (Bag-REPTree) [30] are also popular.…”