Pairwise Ranking Loss for Multi-Task Learning in Recommender Systems

Multi-Task Learning (MTL) plays a crucial role in real-world advertising applications such as recommender systems, aiming to achieve robust representations while minimizing resource consumption. MTL endeavors to simultaneously optimize multiple tasks to construct a unified model serving diverse objectives. In online advertising systems, tasks like Click-Through Rate (CTR) and Conversion Rate (CVR) are often treated as MTL problems concurrently. However, it has been overlooked that a conversion ($y{cvr}=1$) necessitates a preceding click ($y{ctr}=1$). In other words, while certain CTR tasks are associated with corresponding conversions, others lack such associations. Moreover, the likelihood of noise is significantly higher in CTR tasks where conversions do not occur compared to those where they do, and existing methods lack the ability to differentiate between these two scenarios. In this study, exposure labels corresponding to conversions are regarded as definitive indicators, and a novel task-specific loss is introduced by calculating a \textbf{p}air\textbf{wise} \textbf{r}anking (PWiseR) loss between model predictions, manifesting as pairwise ranking loss, to encourage the model to rely more on them. To demonstrate the effect of the proposed loss function, experiments were conducted on different MTL and Single-Task Learning (STL) models using four distinct public MTL datasets, namely Alibaba FR, NL, US, and CCP, along with a proprietary industrial dataset. The results indicate that our proposed loss function outperforms the BCE loss function in most cases in terms of the AUC metric.