ADSNet: Cross-Domain LTV Prediction with an Adaptive Siamese Network in Advertising

TL;DR

This work tackles sparse real-world LTV data in advertising by introducing ADSNet, a cross-domain transfer framework with a gain-based selective mechanism and a Domain Adaptation Module to mitigate negative transfer. The method combines a robust LTV backbone (encoding, expert, and tower layers) with a Difference Pseudo-Siamese Network to evaluate information gain from external data, and a domain-bridging adapter to align domain representations. Key contributions include the gain evaluation strategy, an iterative alignment training regime, and strong offline plus online validation showing improvements in Norm GINI and LTV/GMV, especially for long-tail predictions. Practically, ADSNet demonstrates notable improvements on an industrial-scale dataset and in live A/B tests, indicating its potential for robust cross-domain LTV estimation in advertising systems.

Abstract

Advertising platforms have evolved in estimating Lifetime Value (LTV) to better align with advertisers' true performance metric. However, the sparsity of real-world LTV data presents a significant challenge to LTV predictive model(i.e., pLTV), severely limiting the their capabilities. Therefore, we propose to utilize external data, in addition to the internal data of advertising platform, to expand the size of purchase samples and enhance the LTV prediction model of the advertising platform. To tackle the issue of data distribution shift between internal and external platforms, we introduce an Adaptive Difference Siamese Network (ADSNet), which employs cross-domain transfer learning to prevent negative transfer. Specifically, ADSNet is designed to learn information that is beneficial to the target domain. We introduce a gain evaluation strategy to calculate information gain, aiding the model in learning helpful information for the target domain and providing the ability to reject noisy samples, thus avoiding negative transfer. Additionally, we also design a Domain Adaptation Module as a bridge to connect different domains, reduce the distribution distance between them, and enhance the consistency of representation space distribution. We conduct extensive offline experiments and online A/B tests on a real advertising platform. Our proposed ADSNet method outperforms other methods, improving GINI by 2$\%$. The ablation study highlights the importance of the gain evaluation strategy in negative gain sample rejection and improving model performance. Additionally, ADSNet significantly improves long-tail prediction. The online A/B tests confirm ADSNet's efficacy, increasing online LTV by 3.47$\%$ and GMV by 3.89$\%$.

Figures (5)

• Figure 1: Illustration of the advertising system conversion funnel and challenges, including: (a) the sparsity of internal purchase data, (b) introducing external data, (c) negative transfer due to data distribution shift.
• Figure 2: (a) Conventional multi-domain models incorporate source domain knowledge by simply aggregating both source and target domain data and training in a joint manner, which result in introducing noisy samples and even lead to the problem of negative transfer due to data distribution shifts. (b) Our ADSNet explores a cross-domain transfer learning method. It employs the pseudo-siamese network to differentially evaluate the information gain provided by the source domain, which supports the rejection of negative gain samples, thereby helping the model learn information that is beneficial to the target domain.
• Figure 3: Overview of our proposed ADSNet approach. A pseudo-siamese architecture (part 1) is employed to establish a metric to contrast the differences between two networks, allowing the calculation of each input's gain and supporting the rejection of negative gain samples (part 2). A domain adaption module (part 3) is proposed to promote consistency across different domains.
• Figure 4: Tendency of negative gain rejection rate and GINI during training $w/$or $w/o$ the Gain Evaluation Strategy (GES).
• Figure 5: Comparison over the interval of sample size.