Dynamic Network-Based Two-Stage Time Series Forecasting for Affiliate Marketing
Zhe Wang, Yaming Yang, Ziyu Guan, Bin Tong, Rui Wang, Wei Zhao, Hongbo Deng
TL;DR
The paper tackles promoter attribution in affiliate marketing by introducing propagation scale, the sum of sales across a promoter's descendant promotion paths, as a measure of indirect influence $y(m)=\sum_{o\in\mathcal{O}(m)}\phi(o)$. It proposes a two-stage DNTS framework that first forecasts self-sales and promotion network structures, then synthesizes the propagation scale using differentiable descendant sampling and a matrix-aggregation mechanism, decoupling dynamic network evolution from signal modeling. The model combines an inception-based Temporal Convolution Module for self-sales, a Local Spatial Module with DFS-based descendant sampling and an inductive GCN, and a Global Hypergraph Module to capture cross-item interactions, followed by a decoding phase using Gumbel-Softmax and activation ratios with a multi-term loss $\mathcal{L}=\mathcal{L}_{main}+\mathcal{L}_{aux}$. To address data sparsity and volatility, the approach employs item-specific promoter sub-tables and promoter-activation supervision, along with auxiliary self-sales supervision. Offline experiments show DNTS outperforms strong baselines, and online A/B tests on the Alimama platform yield GMV gains of $+9.29\%$ and sales gains of $+5.89\%$, demonstrating clear practical impact for affiliate-marketing optimization.
Abstract
In recent years, affiliate marketing has emerged as a revenue-sharing strategy where merchants collaborate with promoters to promote their products. It not only increases product exposure but also allows promoters to earn a commission. This paper addresses the pivotal yet under-explored challenge in affiliate marketing: accurately assessing and predicting the contributions of promoters in product promotion. We design a novel metric for evaluating the indirect contributions of the promoter, called propagation scale. Unfortunately, existing time series forecasting techniques fail to deliver accurate predictions due to the propagation scale being influenced by multiple factors and the inherent complexities arising from dynamic scenarios. To address this issue, we decouple the network structure from the node signals and propose a two-stage solution: initially, the basic self-sales and network structure prediction are conducted separately, followed by the synthesis of the propagation scale. Specifically, we design a graph convolution encoding scheme based on descendant neighbors and incorporate hypergraph convolution to efficiently capture complex promotional dynamics. Additionally, three auxiliary tasks are employed: self-sales prediction for base estimations, descendant prediction to synthesize propagation scale, and promoter activation prediction to mitigate high volatility issues. Extensive offline experiments on large-scale industrial datasets validate the superiority of our method. We further deploy our model on Alimama platform with over $100,000$ promoters, achieving a $9.29\%$ improvement in GMV and a $5.89\%$ increase in sales volume.