Cleanformer: A Confident Learning Based ERP Label Denoising Framework for Public Attitude Assessment to Recycled Water

Water scarcity is driving the global adoption of recycled water as an eco-friendly and sustainable solution. Understanding the public’s implicit attitudes, which often diverge from explicit attitudes, is crucial before initiating recycled water programs. In this paper, we proposed a deep learning-based framework to assess the discrepancies between participants’ explicit and implicit attitudes. The results revealed widespread discrepancies between the public’s implicit and explicit attitudes towards recycled water, with the public tending to exhibit more negative implicit attitudes toward recycled water. More than one-third of the samples had discrepancies between implicit and explicit attitudes, and among these samples, the largest number reported their explicit attitudes as neutral. We also discovered that among these neutral samples, 66.15% exhibited negative implicit attitudes. Under these conditions, our model achieved a classification accuracy of 88.04% for the three-class attitude classification and 75.54% for the five-class attitude classification. These results suggested that more attention needs to be given to assessing the public’s implicit attitudes in the promotion of recycled water to capture their true attitudes accurately. Additionally, the proposed method provides a new perspective for the attitude assessment of sustainable productions. Further research on additional demographic factors is still needed to explore more generalizable results.