Semantic Context Information Modeling With Neural Networks in Customer Order Behavior Classification

Demand planning in the semiconductor industry can be complicated due to challenges such as extended cycle times, rapid innovation cycles, and the Bullwhip Effect. Approaches that provide a deeper understanding of customer orders and their associated demand are crucial to enhance demand planning accuracy. Previous studies have employed convolutional neural networks (CNNs) on heat map representations of customer order transactions to effectively classify customer order behaviors (COBs), leading to improved insights into customer behavior. However, these approaches have primarily focused on analyzing customer order patterns without considering contextual information, such as financial or market-related data, which can benefit the classification process. Therefore, we propose a Semantic Context Information Modeling methodology for Neural Networks (SCIM-NN) based on ontologies, knowledge graph embeddings, and multi-stream neural networks to include context information for a classification task. We show the application of SCIM-NN on a use case in the domain of COB and evaluate the performance of the context-aware model on customer data of Infineon Technologies AG. Results indicate that including context information improves the overall classification performance compared to a benchmark CNN.

Automated summary

Demand planning in the semiconductor industry is challenging due to extended cycle times, rapid innovation cycles, and the Bullwhip Effect. This study proposes a methodology called SCIM-NN that incorporates semantic context information into the classification task, resulting in improved overall performance compared to a benchmark CNN model. The application of SCIM-NN on a use case in the domain of COB demonstrates its effectiveness on customer data of Infineon Technologies AG.