E-commerce consumption path prediction and loss early warning system based on multi-modal fusion
By fusing the Transformer-XL model with multimodal data, the difficulty of identifying the state transition phase in e-commerce user behavior analysis was solved, enabling accurate prediction of user states and timely adjustment of recommendation signals, thereby improving the operational efficiency and user experience of e-commerce platforms.
Patent Information
- Application Number
- CN202511501042.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-21
- Publication Date
- 2026-01-13
AI Technical Summary
In existing e-commerce user behavior analysis and decision path modeling, the lack of effective integration of multimodal data leads to difficulties in identifying the transition phase of user state transformation, vague or delayed user state prediction, inaccurate recommendation signal output, biased consumption path construction, and untimely churn warning.
The Transformer-XL model is used to train sub-units. The model is trained by multimodal fusion feature sequences. Combined with e-commerce log processing and segmentation length optimization sub-units, a multimodal fusion modeling and state prediction module, a path construction and recommendation signal generation module, and a state optimization and recommendation signal adjustment module are constructed to achieve effective fusion of multimodal data and state prediction.
It improved the accuracy of user status prediction, optimized the construction of consumption paths, enhanced the timeliness and accuracy of recommendation signals, reduced the risk of user churn, and improved the operational efficiency and user experience of e-commerce platforms.
Smart Images

Figure CN121329488A_ABST