Product deletion, as an important part of product management, can timely release productivity and reduce the impact of suboptimal products on consumer satisfaction. An informed product deletion decision is based on the observation of the temporal changes in multitype features. However, with increasing market volatility, traditional product deletion decision-making methods are becoming less effective and more inefficient for highly dynamic market environments. Thus, this paper proposes a novel AI-based deep neural network architecture to address the increasingly complex product deletion decision-making problem. First, we have developed three sub-models to capture local and global patterns from heterogeneous information: namely consumer, market, and consumer-market sub-model. Each sub-model incorporates a convolutional neural network (CNN), a recurrent-skip neural network (RSNN), and a gated recurrent unit (GRU), which respectively extract non-linear, long-term, and short-term temporal dependencies. To address the challenge of deep learning in capturing fixed-scale changes between multivariate time series (MTS) and the decision target, we additionally design an automatic logistic regression component running in parallel with the three sub-models. Through a series of experiments, the effectiveness and robustness of our proposed approach have been validated in product deletion decision-making. It demonstrates superior performance compared to conventional machine learning and time series classification algorithms. In general, this paper not only offers potential business value for product deletion decision-making, but also provides a reference framework for similar complex prediction problems in interdisciplinary fields.
Liang et al. (Tue,) studied this question.