Moyu Zhang

Moyu Zhang, Xinning Zhu, Chunhong Zhang et al.

As the core of the Knowledge Tracking (KT) task, assessing students' dynamic mastery of knowledge concepts is crucial for both offline teaching and online educational applications. Since students' mastery of knowledge concepts is often unlabeled, existing KT methods rely on the implicit paradigm of historical practice to mastery of knowledge concepts to students' responses to practices to address the challenge of unlabeled concept mastery. However, purely predicting student responses without imposing specific constraints on hidden concept mastery values does not guarantee the accuracy of these intermediate values as concept mastery values. To address this issue, we propose a principled approach called Counterfactual Monotonic Knowledge Tracing (CMKT), which builds on the implicit paradigm described above by using a counterfactual assumption to constrain the evolution of students' mastery of knowledge concepts.

Moyu Zhang, Yun Chen, Yujun Jin et al.

Generative pre-training via discrete diffusion provides dense reconstruction supervision across all feature fields simultaneously, mitigating representation collapse from data sparsity in CTR prediction. However, all existing generative CTR methods share a fundamental limitation: the reconstruction objective assigns equal training weight to every feature field, ignoring the profound heterogeneity of reconstruction difficulty across high-cardinality ID fields, sparse categorical attributes, numerical values, and behavioral sequences. This causes easy fields to dominate training gradients while the hardest but most informative fields remain chronically underfit, a problem we term the generative difficulty imbalance.We propose HeteGenCTR, which resolves this imbalance through per-field learnable difficulty parameters jointly trained with the denoising network. This unified signal drives two coordinated components without additional hyperparameters: a self-balancing loss that automatically reallocates gradient budget toward harder fields with a provably stable equilibrium, and a difficulty-guided attention mechanism that suppresses the influence of already-converged easy fields while amplifying cross-field information flow toward hard fields. Both components share the same learned signal and remain mutually consistent throughout training. Experiments on five CTR benchmarks and a seven-day online A/B test demonstrate consistent, statistically significant improvements over state-of-the-art baselines, with disproportionate gains for cold-start and long-tail users.

Moyu Zhang, Yun Chen, Yujun Jin et al.

Scaling test-time compute has proven highly effective for language models, yet this opportunity remains largely unexplored for industrial Click-Through Rate (CTR) prediction. CTR models suffer from a fundamental asymmetry: feature combinations well-represented in training yield confident predictions, while sparsely observed ones produce unreliable outputs. Existing training-phase solutions such as adaptive gating learn a fixed selection function subject to the same sparsity, offering no per-instance recourse at deployment.We propose UTTSI (Uncertainty-Triggered Test-Time Selective Inference), a training-free model-agnostic framework that scales inference depth proportionally to per-instance uncertainty. A dual-signal estimator combining model logit confidence with a data-level frequency prior distinguishes epistemic uncertainty from aleatoric ambiguity. Every instance undergoes adaptive feature filtering to remove unreliable embeddings; uncertain instances additionally receive stochastic feature-path explorations whose predictions are aggregated via consistency-weighted ensembling. Confident instances bypass exploration entirely, keeping average overhead at approximately $2.8\times$ base model cost with worst-case latency unchanged.Experiments on four datasets with three backbone architectures demonstrate consistent, statistically significant gains over all training-phase baselines. A seven-day online A/B test further confirms a 5.3% relative CTR gain ($p < 0.01$), establishing selective test-time compute allocation as a practical complement to training-phase advances for CTR prediction.

Moyu Zhang, Xinning Zhu, Chunhong Zhang et al.

Knowledge tracing (KT) aims to predict students' responses to practices based on their historical question-answering behaviors. However, most current KT methods focus on improving overall AUC, leaving ample room for optimization in modeling sequences of excessive or insufficient lengths. As sequences get longer, computational costs will increase exponentially. Therefore, KT methods usually truncate sequences to an acceptable length, which makes it difficult for models on online service systems to capture complete historical practice behaviors of students with too long sequences. Conversely, modeling students with short practice sequences using most KT methods may result in overfitting due to limited observation samples. To address the above limitations, we propose a model called Sequence-Flexible Knowledge Tracing (SFKT).

Yun Chen, Moyu Zhang, Jinxin Hu et al.

Neural Architecture Search (NAS) deployment in industrial production systems faces a fundamental validation bottleneck: verifying a single candidate architecture pi requires evaluating the deployed ensemble of M models, incurring prohibitive O(M) computational cost per candidate. This cost barrier severely limits architecture iteration frequency in real-world applications where ensembles (M=50-200) are standard for robustness. This work introduces Ensemble-Decoupled Architecture Search, a framework that leverages ensemble theory to predict system-level performance from single-learner evaluation. We establish the Ensemble-Decoupled Theory with a sufficient condition for monotonic ensemble improvement under homogeneity assumptions: a candidate architecture pi yields lower ensemble error than the current baseline if rho(pi) < rho(pi_old) - (M / (M - 1)) * (Delta E(pi) / sigma^2(pi)), where Delta E, rho, and sigma^2 are estimable from lightweight dual-learner training. This decouples architecture search from full ensemble training, reducing per-candidate search cost from O(M) to O(1) while maintaining O(M) deployment cost only for validated winners. We unify solution strategies across pipeline continuity: (1) closed-form optimization for tractable continuous pi (exemplified by feature bagging in CTR prediction), (2) constrained differentiable optimization for intractable continuous pi, and (3) LLM-driven search with iterative monotonic acceptance for discrete pi. The framework reveals two orthogonal improvement mechanisms -- base diversity gain and accuracy gain -- providing actionable design principles for industrial-scale NAS. All theoretical derivations are rigorous with detailed proofs deferred to the appendix. Comprehensive empirical validation will be included in the journal extension of this work.

Moyu Zhang, Xinning Zhu, Chunhong Zhang et al.

The Knowledge Tracing (KT) task plays a crucial role in personalized learning, and its purpose is to predict student responses based on their historical practice behavior sequence. However, the KT task suffers from data sparsity, which makes it challenging to learn robust representations for students with few practice records and increases the risk of model overfitting. Therefore, in this paper, we propose a Cognition-Mode Aware Variational Representation Learning Framework (CMVF) that can be directly applied to existing KT methods. Our framework uses a probabilistic model to generate a distribution for each student, accounting for uncertainty in those with limited practice records, and estimate the student's distribution via variational inference (VI). In addition, we also introduce a cognition-mode aware multinomial distribution as prior knowledge that constrains the posterior student distributions learning, so as to ensure that students with similar cognition modes have similar distributions, avoiding overwhelming personalization for students with few practice records. At last, extensive experimental results confirm that CMVF can effectively aid existing KT methods in learning more robust student representations. Our code is available at https://github.com/zmy-9/CMVF.

Moyu Zhang, Yongxiang Tang, Jinxin Hu et al.

Existing methods often adjust representations adaptively only after aggregating user behavior sequences. This coarse-grained approach to re-weighting the entire user sequence hampers the model's ability to accurately model the user interest migration across different scenarios. To enhance the model's capacity to capture user interests from historical behavior sequences in each scenario, we develop a ranking framework named the Scenario-Adaptive Fine-Grained Personalization Network (SFPNet), which designs a kind of fine-grained method for multi-scenario personalized recommendations. Specifically, SFPNet comprises a series of blocks named as Scenario-Tailoring Block, stacked sequentially. Each block initially deploys a parameter personalization unit to integrate scenario information at a coarse-grained level by redefining fundamental features. Subsequently, we consolidate scenario-adaptively adjusted feature representations to serve as context information. By employing residual connection, we incorporate this context into the representation of each historical behavior, allowing for context-aware fine-grained customization of the behavior representations at the scenario-level, which in turn supports scenario-aware user interest modeling.

Yi Xu, Moyu Zhang, Chenxuan Li et al.

Recommender systems traditionally represent items using unique identifiers (ItemIDs), but this approach struggles with large, dynamic item corpora and sparse long-tail data, limiting scalability and generalization. Semantic IDs, derived from multimodal content such as text and images, offer a promising alternative by mapping items into a shared semantic space, enabling knowledge transfer and improving recommendations for new or rare items. However, existing methods face two key challenges: (1) balancing cross-modal synergy with modality-specific uniqueness, and (2) bridging the semantic-behavioral gap, where semantic representations may misalign with actual user preferences. To address these challenges, we propose Multimodal Mixture-of-Quantization (MMQ), a two-stage framework that trains a novel multimodal tokenizer. First, a shared-specific tokenizer leverages a multi-expert architecture with modality-specific and modality-shared experts, using orthogonal regularization to capture comprehensive multimodal information. Second, behavior-aware fine-tuning dynamically adapts semantic IDs to downstream recommendation objectives while preserving modality information through a multimodal reconstruction loss. Extensive offline experiments and online A/B tests demonstrate that MMQ effectively unifies multimodal synergy, specificity, and behavioral adaptation, providing a scalable and versatile solution for both generative retrieval and discriminative ranking tasks.

Yi Xu, Chaofan Fan, Moyu Zhang et al.

Click-through rate (CTR) prediction tasks typically estimate the probability of a user clicking on a candidate item by modeling both user behavior sequence features and the item's contextual features, where the user behavior sequence is particularly critical as it dynamically reflects real-time shifts in user interest. Traditional CTR models often aggregate this dynamic sequence into a single vector before interacting it with contextual features. This approach, however, not only leads to behavior information loss during aggregation but also severely limits the model's capacity to capture interactions between contextual features and specific user behaviors, ultimately impairing its ability to capture fine-grained behavioral details and hindering models' prediction accuracy. Conversely, a naive approach of directly interacting with each user action with contextual features is computationally expensive and introduces significant noise from behaviors irrelevant to the candidate item. This noise tends to overwhelm the valuable signals arising from interactions involving more behaviors relevant to the candidate item. Therefore, to resolve the above issue, we propose a Core-Behaviors and Distributional-Compensation Dual-View Interaction Network (CDNet), which bridges the gap between sequential and contextual feature interactions from two complementary angles: a fine-grained interaction involving the most relevant behaviors and contextual features, and a coarse-grained interaction that models the user's overall interest distribution against the contextual features. By simultaneously capturing important behavioral details without forgoing the holistic user interest, CDNet effectively models the interplay between sequential and contextual features without imposing a significant computational burden. Ultimately, extensive experiments validate the effectiveness of CDNet.

Yi Xu, Moyu Zhang, Chaofan Fan et al.

Click-through rate (CTR) prediction models estimates the probability of a user-item click by modeling interactions across a vast feature space. A fundamental yet often overlooked challenge is the inherent heterogeneity of these features: their sparsity and information content vary dramatically. For instance, categorical features like item IDs are extremely sparse, whereas numerical features like item price are relatively dense. Prevailing CTR models have largely ignored this heterogeneity, employing a uniform feature interaction strategy that inputs all features into the interaction layers simultaneously. This approach is suboptimal, as the premature introduction of low-information features can inject significant noise and mask the signals from information-rich features, which leads to model collapse and hinders the learning of robust representations. To address the above challenge, we propose a Multi-Granularity Information-Aware Deferred Interaction Network (MGDIN), which adaptively defers the introduction of features into the feature interaction process. MGDIN's core mechanism operates in two stages: First, it employs a multi-granularity feature grouping strategy to partition the raw features into distinct groups with more homogeneous information density in different granularities, thereby mitigating the effects of extreme individual feature sparsity and enabling the model to capture feature interactions from diverse perspectives. Second, a delayed interaction mechanism is implemented through a hierarchical masking strategy, which governs when and how each group participates by masking low-information groups in the early layers and progressively unmasking them as the network deepens. This deferred introduction allows the model to establish a robust understanding based on high-information features before gradually incorporating sparser information from other groups...

Moyu Zhang, Yun Chen, Yujun Jin et al.

Recently, a growing body of research has focused on either optimizing CTR model architectures to better model feature interactions or refining training objectives to aid parameter learning, thereby achieving better predictive performance. However, previous efforts have primarily focused on the training phase, largely neglecting opportunities for optimization during the inference phase. Infrequently occurring feature combinations, in particular, can degrade prediction performance, leading to unreliable or low-confidence outputs. To unlock the predictive potential of trained CTR models, we propose a Model-Agnostic Test-Time paradigm (MATT), which leverages the confidence scores of feature combinations to guide the generation of multiple inference paths, thereby mitigating the influence of low-confidence features on the final prediction. Specifically, to quantify the confidence of feature combinations, we introduce a hierarchical probabilistic hashing method to estimate the occurrence frequencies of feature combinations at various orders, which serve as their corresponding confidence scores. Then, using the confidence scores as sampling probabilities, we generate multiple instance-specific inference paths through iterative sampling and subsequently aggregate the prediction scores from multiple paths to conduct robust predictions. Finally, extensive offline experiments and online A/B tests strongly validate the compatibility and effectiveness of MATT across existing CTR models.

Moyu Zhang, Xinning Zhu, Chunhong Zhang et al.

With the increasing demands of personalized learning, knowledge tracing has become important which traces students' knowledge states based on their historical practices. Factor analysis methods mainly use two kinds of factors which are separately related to students and questions to model students' knowledge states. These methods use the total number of attempts of students to model students' learning progress and hardly highlight the impact of the most recent relevant practices. Besides, current factor analysis methods ignore rich information contained in questions. In this paper, we propose Multi-Factors Aware Dual-Attentional model (MF-DAKT) which enriches question representations and utilizes multiple factors to model students' learning progress based on a dual-attentional mechanism. More specifically, we propose a novel student-related factor which records the most recent attempts on relevant concepts of students to highlight the impact of recent exercises. To enrich questions representations, we use a pre-training method to incorporate two kinds of question information including questions' relation and difficulty level. We also add a regularization term about questions' difficulty level to restrict pre-trained question representations to fine-tuning during the process of predicting students' performance. Moreover, we apply a dual-attentional mechanism to differentiate contributions of factors and factor interactions to final prediction in different practice records. At last, we conduct experiments on several real-world datasets and results show that MF-DAKT can outperform existing knowledge tracing methods. We also conduct several studies to validate the effects of each component of MF-DAKT.