Expert-sharing patterns have emerged as promising paradigms for multi-task learning (MTL) in recommender systems, enabling efficient resource allocation and dynamic modeling of diverse tasks. In this paper, we observe that high-gating (leader) and low-gating (auxiliary) experts play distinct roles in MTL: leader experts dominate task-specific predictions, while auxiliary experts, despite their lower gating scores, often contain complementary knowledge that can enhance model performance. However, critical challenges persist: how to effectively identify and utilize the knowledge of the leader and auxiliary experts in a synergistic manner? To address this, we propose a novel Dynamic Experts Synergy (DES) mechanism that integrates Entropy-driven Experts Classification (EEC) and Multi-view Knowledge Recycle (MVKR). EEC dynamically partitions experts into leader and auxiliary groups by analyzing task-specific prediction and gating entropy, enabling adaptive allocation aligned with real-time task difficulty. MVKR effectively revisits knowledge from auxiliary experts through utility, diversity, and task-relatedness perspectives, ensuring comprehensive knowledge utilization. Extensive experiments on five datasets demonstrate the superiority of our DES against state-of-the-art methods.

Multi-task learning (MTL) is a widely adopted machine learning paradigm in recommender systems. However, existing MTL models often suffer from performance degeneration with negative transfer and seesaw phenomena. Some works attempt to alleviate the negative transfer and seesaw issues by separating task-specific and shared experts to mitigate the harmful interference between task-specific and shared knowledge. Despite the success of these efforts, task-specific and shared knowledge have still not been thoroughly decoupled. There may still exist unnecessary mixture between the shared and task-specific knowledge, which may harm MLT models' performances. To tackle this problem, in this paper, we propose multi-task knowledge disentanglement network (MTKDN) to further reduce harmful interference between the shared and task-specific knowledge. Specifically, we propose a novel contrastive disentanglement mechanism to explicitly decouple the shared and task-specific knowledge in corresponding hidden spaces. In this way, the unnecessary mixture between shared and task-specific knowledge can be reduced. As for optimization objectives, we propose individual optimization objectives for shared and task-specific experts, by which we can encourage these two kinds of experts to focus more on extracting the shared and task-specific knowledge, respectively. Additionally, we propose a margin regularization to ensure that the fusion of shared and task-specific knowledge can outperform exploiting either of them alone. We conduct extensive experiments on open-source large-scale recommendation datasets. The experimental results demonstrate that MTKDN significantly outperforms state-of-the-art MTL models. In addition, the ablation experiments further verify the necessity of our proposed contrastive disentanglement mechanism and the novel loss settings.

As a milestone research combining recommender systems and knowledge graphs (KG), Knowledge Graph Attention Network (KGAT) has achieved great success in the field of recommender systems by proposing a new approach that implements explicit end-to-end modeling of higher-order relationships in a graph neural network framework to provide better item-assisted information recommendations. A series of methods following KGAT provide more solutions for KG-based recommendations. However, over-reliance on high-quality knowledge graphs is a bottleneck for such methods. Specifically, the long-tailed distribution of entities of KG and noise issues in the real world will make item-entity dependent relations deviate from reflecting true characteristics and significantly harm the performance of modeling user preference. Contrastive learning, as a novel method that is employed for data augmentation and denoising, provides inspiration to fill this research gap. However, the mainstream work only focuses on the long-tail properties of the number of items clicked, while ignoring that the long-tail properties of total number of clicks per user may also affect the performance of the recommendation model. Therefore, to tackle these problems, motivated by the Debiased Contrastive Learning of Unsupervised Sentence Representations (DCLR), we propose Two-Level Debiased Contrastive Graph Learning (TDCGL) model. Specifically, we design the Two-Level Debiased Contrastive Learning (TDCL) and deploy it in the KG, which is conducted not only on User-Item pairs but also on User-User pairs for modeling higher-order relations. Also, to reduce the bias caused by random sampling in contrastive learning, with the exception of the negative samples obtained by random sampling, we add a noise-based generation of negation to ensure spatial uniformity. Considerable experiments on open-source datasets demonstrate that our method has excellent anti-noise capability and significantly outperforms state-of-the-art baselines. In addition, ablation studies about the necessity for each level of TDCL are conducted.

Recently, multi-task learning based on the deep neural network has been successfully applied in many recommender system scenarios. The prediction quality of current mainstream multi-task models often relies on the extent to which the relationships among tasks are extracted. Much of the prior research work has focused on two important tasks in recommender systems: predicting click-through rate (CTR) and post-click conversion rate (CVR), which rely on sequential user action pattern of impression → click → conversion. Therefore, there exists sequential dependence between CTR and CVR tasks. However, there is no satisfactory solution to explicitly model the sequential dependence among tasks without sacrificing the first task in terms of the design of the model network structure. In this paper, inspired by the Multi-task Network Cascades (MNC) and Adaptive Information Transfer Multi-task (AITM) frameworks, we propose a Multi-level Network Cascades Model (MNCM) based on the pattern of specific and shared experts separation. In MNCM, we introduce two types of information transfer modules: Task-Level Information Transfer Module (TITM) and Expert-Level Information Transfer Module (EITM), which can learn transferred information adaptively from task level and task-specific experts level, respectively, thereby fully capture sequential dependence among tasks. Compared with AITM, MNCM effectively avoids the problem of the first task in a task sequence becoming the sacrificial side of the seesaw phenomenon and contributes to mitigating potential conflicts among tasks. We conduct considerable experiments based on open-source large-scale recommendation datasets. The experimental results demonstrate that MNCM outperforms AITM and the mainstream baseline models in the mixture-experts-bottom pattern and probability-transfer pattern. In addition, we conduct an ablation study on the necessity of introducing two kinds of information transfer modules and verify the effectiveness of this pattern.

In recent years, multi-task learning models based on deep learning in recommender systems have attracted increasing attention from researchers in industry and academia. Accurately estimating post-click conversion rate (CVR) is often considered as the primary task of multi-task learning in recommender systems. However, some advertisers may try to get higher click-through rates (CTR) by over-decorating their ads, which may result in excessive exposure to samples with lower CVR. For example, some only eye-catching clickbait have higher CTR, but actually, CVR is very low. As a result, the overall performance of the recommender system will be hurt. In this paper, we introduce a novelty auxiliary task called CTnoCVR, which aims to predict the probability of events with click but no-conversion, in various state-of-the-art multi-task models of recommender systems to promote samples with high CVR but low CTR. Plentiful Experiments on a large-scale dataset gathered from traffic logs of Taobao's recommender system demonstrate that the introduction of CTnoCVR task significantly improves the prediction effect of CVR under various multi-task frameworks. In addition, we conduct the online test and evaluate the effectiveness of our proposed method to make those samples with high CVR and low CTR rank higher.