Abstract:Multi-touch attribution (MTA) currently plays a pivotal role in achieving a fair estimation of the contributions of each advertising touchpoint to-wards conversion behavior, deeply influencing budget allocation and advertising recommenda-tion. Previous works attempted to eliminate the bias caused by user preferences to achieve the unbiased assumption of the conversion model. The multi-model collaboration method is not ef-ficient, and the complete elimination of user in-fluence also eliminates the causal effect of user features on conversion, resulting in limited per-formance of the conversion model. This paper re-defines the causal effect of user features on con-versions and proposes a novel end-to-end ap-proach, Deep Causal Representation for MTA (DCRMTA). Our model focuses on extracting causa features between conversions and users while eliminating confounding variables. Fur-thermore, extensive experiments demonstrate DCRMTA's superior performance in converting prediction across varying data distributions, while also effectively attributing value across dif-ferent advertising channels.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the causal relationship modeling and bias elimination in Multi - Touch Attribution (MTA). Specifically, MTA aims to fairly evaluate the contribution of each advertising touchpoint to user conversion behavior, thereby influencing budget allocation and advertising recommendations. However, existing methods have limitations in dealing with biases caused by user preferences. Completely eliminating user influence will also lose the causal effect of user characteristics on conversion, which limits the model performance. To this end, the paper redefines the causal effect of user characteristics on conversion and proposes a new end - to - end method - Deep Causal Representation for MTA (DCRMTA). This model focuses on extracting the causal characteristics between conversion and users while eliminating confounding variables. Through extensive experiments, the superior performance of DCRMTA in conversion prediction and advertising channel value allocation under different data distributions has been proven. ### Main problem summary: 1. **Limitations of existing methods**: - The multi - model collaboration method is inefficient. - Completely eliminating user influence will lose the causal effect of user characteristics on conversion, resulting in limited model performance. 2. **Challenges in causal relationship modeling**: - The influence of user characteristics at different time points is complex, and it is impossible to observe successful and failed conversion results simultaneously, making it difficult to obtain independent causal effects. 3. **Data sparsity and selection bias**: - Sparse touchpoint data and user selection bias make it difficult to learn user characteristics. - The problem of missing data (MNAR) further exacerbates this challenge. ### Solutions: - **DCRMTA model**: - Extract causal features: Deeply extract static features in the user - advertising interaction history through the Causal Attention Module. - Eliminate confounding variables: Combine the Gradient Reversal Layer (GRL) and the LSTM - based counterfactual model to obtain dynamic features from sequential advertising browsing data. - Conversion prediction: Calculate the importance weight of each advertisement through joint game theory to optimize advertising expenditure. ### Formula representation: 1. **Loss function**: \[ \mathcal{L}_{\text{total}}=\mathcal{L}_{\text{cpred}}+\mathcal{L}_{\text{rev}} \] where, \[ \mathcal{L}_{\text{cpred}}=\beta \sum_{i = 1}^{N}\text{CE}(p_i,y_i)+\gamma \sum_{i = 1}^{N}\text{CE}(\hat{p}_i,y_i) \] \[ \mathcal{L}_{\text{rev}}=\alpha \sum_{i = 1}^{N}\sum_{t = 1}^{T}\text{CE}(c_{t}^{\text{rev}},c_t) \] 2. **Shapley value calculation**: \[ SV_{t_i}=\sum_{S\subseteq J_i\setminus \{tp_{t_i}\}}\frac{|S|!(|J_i|-|S|- 1)!}{|J_i|!}[f(S\cup \{c_{t_i}\})-f(S)] \] Normalized Shapley value: \[ \text{attr}_{t_i}=\frac{(SV_{t_i})^+}{\sum_{t = 1}^{T_i}(SV_{t_i})^+} \] Through these methods, DCRMTA can more accurately evaluate the contribution of each advertising touchpoint to user conversion, thereby optimizing the advertising placement strategy.

DCRMTA: Unbiased Causal Representation for Multi-touch Attribution

CausalMTA: Eliminating the User Confounding Bias for Causal Multi-touch Attribution

Interpretable Deep Learning Model for Online Multi-touch Attribution

CAMTA: Causal Attention Model for Multi-touch Attribution

Causally Driven Incremental Multi Touch Attribution Using a Recurrent Neural Network

Deep Neural Net with Attention for Multi-channel Multi-touch Attribution

Learning Multi-touch Conversion Attribution with Dual-attention Mechanisms for Online Advertising

DCMT: A Direct Entire-Space Causal Multi-Task Framework for Post-Click Conversion Estimation.

A Graphical Point Process Framework for Understanding Removal Effects in Multi-Touch Attribution

Large-scale Causal Approaches to Debiasing Post-click Conversion Rate Estimation with Multi-task Learning

Representation Learning-Assisted Click-Through Rate Prediction

Modeling the Sequential Dependence among Audience Multi-step Conversions with Multi-task Learning in Targeted Display Advertising.

Bayesian Modeling of Marketing Attribution

Sim-to-Real Causal Transfer: A Metric Learning Approach to Causally-Aware Interaction Representations

Deep treatment-adaptive network for causal inference

A decision-making characteristics framework for marketing attribution in practice: Improving empirical procedures

Modeling Multi-Channel Advertising Attribution Across Competitors

Disentangled Latent Representation Learning for Tackling the Confounding M-Bias Problem in Causal Inference

TMH: Two-Tower Multi-Head Attention neural network for CTR prediction

A New Approach for Mobile Advertising Click-Through Rate Estimation Based on Deep Belief Nets

CausalMMM: Learning Causal Structure for Marketing Mix Modeling