669 / 2024-05-07 09:58:10

Predicting Sales in Live Streaming Rooms by Incorporating Language Strategies: A Model Based on Graph Neural Networks

Live Streaming,deep learning,graph neural network,language strategy,sales prediction

This study investigates how to utilize anchor language strategies to predict sales in live e-commerce environments. With the rapid development of live e-commerce, accurate sales forecasting is crucial for business strategies, inventory management, and supply chain operations. Traditional sales forecasting methods rely mainly on historical data and market trends, but the real-time and interactive nature of live e-commerce makes sales data more dynamic and uncertain, making it challenging for traditional models to capture sales trends. Therefore, this study proposes a model based on graph neural networks called Live-Graphormer, which analyzes the graph structure of anchor language strategies, explores node interactions and global structural information, and integrates multidimensional live performance data to complete live sales forecasting.



The study first categorizes anchor language strategies and treats strategy types as nodes in the graph structure, using directed edges to represent connections between strategies. Then, by constructing a graph of anchor language strategies, it analyzes the relationships between different strategy types and their impact on sales. Experiments were conducted using Douyin e-commerce datasets for validation. The results indicate a positive correlation between the number of language strategy types used by anchors and sales, while frequent transitions between strategy types may lead to information overload and have a negative impact on sales. The Live-Graphormer model demonstrated high prediction accuracy in the experiments, providing practical decision support for live e-commerce operations.



Furthermore, the study reviews existing literature on language strategies, sales forecasting, and the application of graphs in text sequences, and provides detailed descriptions of data collection, processing, and model structure. Through parameter settings and ablation experiments, the effectiveness and stability of the model were further validated. Ultimately, this research not only offers new perspectives and methods for sales forecasting in live e-commerce but also provides theoretical support for understanding the role of anchor language strategies in live e-commerce.