Automatic Marketing Theme and Commodity Construction System for E-commerce

When consumers have focused shopping needs, they are more interested in collections of products aligned with specific marketing themes. Therefore, mining marketing themes and their corresponding product collections can help customers save on shopping costs and improve user clicks and purchases within the recommendation system. However, the current system relies on experts to write marketing themes and select relevant products, which suffers from difficulties in mass production, poor timeliness, and low online indicators. Therefore, we propose an automatic system for marketing theme and product construction. This system can automatically generate popular marketing themes and select relevant products, while also improving the online effectiveness of these themes within the recommendation system. Specifically, we first utilize a pretrained language model to generate the marketing themes. Then, we use the theme-commodity consistency module to select the relevant products for the generated themes. Additionally, we build an indicator simulator to evaluate the effectiveness of the generated themes. When the indicator is lower, the selected products are input into the theme-rewriter module to generate more efficient marketing themes. Finally, we employ human screening to ensure system quality control. Both offline experiments and online A/B tests demonstrate the superior performance of our proposed system compared to state-of-the-art methods.