A hybrid LSTM and transformer model for analyzing e-commerce user behavior and path optimization

In the e-commerce live streaming environment, consumer behavior is highly dynamic and uncertain, coupled with drastic fluctuations in traffic, making accurate demand forecasting a major challenge that traditional forecasting models are unable to handle. This study innovatively developed a hybrid model that integrates Long Short-Term Memory (LSTM) network and Transformer, aiming to significantly improve the accuracy of consumer behavior prediction and provide scientific basis for flexible adaptation and efficient operation of the supply chain. This model deeply explores the temporal evolution patterns of consumer behavior through LSTM, and utilizes the powerful global attention mechanism of Transformer to reveal the intricate correlations between behaviors, achieving accurate characterization and forward-looking prediction of consumer behavior in e-commerce live streaming scenarios. Experimental data shows that the LSTM-Transformer model continues to lead in consumer behavior hit rate, far surpassing similar models such as Bidirectional Encoder Representations from Transformers (BERT) Bidirectional Long Short-Term Memory (BiLSTM) and Multi-Entity Multimodal Fusion (MEMF), demonstrating its excellent predictive performance. Further research has confirmed that the deep integration of LSTM and Transformer is the core of model performance improvement. The synergistic effect of the two greatly enhances the ability to analyze and predict complex and changing consumer behavior. This study not only opens up a new path for consumer behavior prediction in the field of e-commerce live streaming, but also provides valuable practice for intelligent optimization of supply chains, with profound theoretical significance and practical application value.