An Adaptive Bypass-Based LDL Method for Marketing Strategy Assessment

In the era of digital marketing, the ability to effectively assess consumer attitudes is crucial for brands to optimize their strategies. However, current methods still heavily rely on traditional psychological evaluations, which lack the precision needed to navigate complex consumer sentiments. Artificial intelligence (AI) now plays a key role in driving the analysis of consumer data, especially for "me-too" brands that adopt imitation strategies. To address these challenges, we explore the task of predicting Consumer Attitude Distribution (CAD) and model it as a Label Distribution Learning (LDL) problem. A significant challenge in applying LDL to predict CAD is effectively modeling and leveraging dependencies among labels. To address this, we propose a novel AI-driven LDL method based on an adaptive bypass mechanism, termed LDL-CAD. Our proposed method, LDL-CAD, utilizes a feedforward neural network (FFN) as the main output component and designs an adaptive bypass mechanism through graph learning. This adaptive bypass learns high-order dependency information among multiple labels in the label distribution space in a data-driven manner, employing graph structures to represent and quantify the strength of dependencies among labels. We validate the effectiveness of LDL-CAD through quantitative and qualitative experiments on marketing datasets. The results show that our LDL-CAD achieves state-of-the-art performance, providing a powerful tool for brands to enhance the precision and effectiveness of their marketing strategies.