Recent advances in large language models (LLMs) have enabled promising progress in diagnosis prediction from electronic health records (EHRs). However, existing LLM-based approaches tend to overfit to historically observed diagnoses, often overlooking novel yet clinically important conditions that are critical for early intervention. To address this, we propose EviCare, an in-context reasoning framework that integrates deep model guidance into LLM-based diagnosis prediction. Rather than prompting LLMs directly with raw EHR inputs, EviCare performs (1) deep model inference for candidate selection, (2) evidential prioritization for set-based EHRs, and (3) relational evidence construction for novel diagnosis prediction. These signals are then composed into an adaptive in-context prompt to guide LLM reasoning in an accurate and interpretable manner. Extensive experiments on two real-world EHR benchmarks (MIMIC-III and MIMIC-IV) demonstrate that EviCare achieves significant performance gains, which consistently outperforms both LLM-only and deep model-only baselines by an average of 20.65% across precision and accuracy metrics. The improvements are particularly notable in challenging novel diagnosis prediction, yielding average improvements of 30.97%.

Medical concepts, the core entities in Electronic Health Records (EHRs), provide essential inputs for clinical decision-making systems. However, most existing healthcare models still rely on massive concept-specific embedding tables, resulting in substantial memory overhead. Recent studies compress medical concepts into discrete code sequences for memory efficiency, but their flat semantic quantization fails to explicitly encode the hierarchical structure of medical ontologies, thereby limiting clinical interpretability. To this end, we propose MedRQ, an ontology-driven residual vector quantization framework that aligns discrete codes with multi-level clinical ontologies. By incorporating hierarchical supervision into the quantization process, MedRQ generates compact and ontology-consistent concept representations that generalize seamlessly across healthcare prediction tasks. Experiments on two real-world EHR datasets demonstrate that MedRQ significantly outperforms state-of-the-art baselines while reducing memory usage.

Recent advances in clinical prediction leverage large language models (LLMs) to extract semantic information from Electronic Health Records (EHRs). However, LLMs could produce biased or hallucinated responses camouflaged by their fluency and realistic appearance, which is unacceptable in diagnosis prediction. Uncertainty estimation (UE) has emerged as an effective approach to address this challenge by quantifying hallucination levels and prediction confidence in LLM outputs. Yet, directly determining diagnosis predictions based on UE remains insufficient, as diagnoses with high uncertainty may still correspond to correct outcomes. To this end, we propose ULoR, an uncertainty-aware leave-one-out refinement framework for reliable diagnosis prediction. Specifically, we first compute the UE scores by integrating statistical information from multiple samples of the model's diagnosis ranking distributions and leverage these scores for initial predictions. Then, guided by the leave-one-out strategy, we construct multiple-choice tasks for high-uncertainty diagnoses using external syndrome knowledge and fine-tune the refinement component to resolve them, thereby confirming or replacing uncertain predictions. Extensive experiments on two real-world EHR datasets demonstrate that ULoR consistently outperforms state-of-the-art baselines, showcasing its practical utility in real-world clinical settings.

Hash-based collaborative filtering (Hash-CF) approaches recently employ efficient Hamming distance of learned binary representations to accelerate recommendations. Benefiting from its probabilistic nature, Variational Autoencoder (VAE) enables robust Hash-CF with stronger generalization ability. However, VAE-based Hash-CF still faces two challenging problems: 1) Traditional VAE urges the latent variables of different users (or items) to fit a unified and monotonous prior distribution, and lacks considerations for distinctive characteristics of users (or items). The obtained representations of users and items with slight individual differentiation may further weaken the performance of Hash-CF for subsequent personalized recommendations. 2) Hash-CF under the VAE framework requires discrete optimization on latent Bernoulli distributions, which are discrete and NP-hard to optimize. In this paper, we propose a Dual Discrete Collaborative Filtering (DDCF) approach, including a cluster-enhanced representation generation module and a CNF-enabled discrete optimization module. The former module mainly develops cluster-aware latent space to generate discriminative representations for users or items with significantly different characteristics. The latter module employs Continuous Normalizing Flow (CNF) to achieve discrete optimization on latent Bernoulli distributions steadily and effectively. Extensive experiments conducted on multiple real-world datasets demonstrate the superiority of our DDCF compared with the state-of-art methods in terms of effectiveness and efficiency.

Accurate user profiles are crucial for personalized recommendation systems to mitigate information overload on large-scale online platforms. While recent advances in large language models have enhanced semantic understanding for profile construction through textual artifacts, existing methods often neglect the higher-order structural patterns inherent in user-item interaction graphs-a key limitation for achieving accurate and diverse recommendations. In this paper, we propose SSPRec, a Higher-order Structure and Semantics-enhanced User Profiling for Recommendation. Specifically, we first introduce a multi-hop proximity matrix over item-item transitions, followed by low-rank approximation and clustering to group users based on behavioral similarity. Group-level user profiles are then distilled via representative keywords extracted from co-interacted items, and collaborative embeddings are concurrently learned from the interaction graph. To integrate collaborative signals with language-based profiles, we introduce a cross-view contrastive objective that encourages coherence between structural and semantic representations. Final recommendations are made using a fused user-item similarity score. Extensive experiments on four real-world datasets show that SSPRec not only outperforms baselines in accuracy (with 46.35% improvements), but also remains diverse and robust, even under incomplete interactions.

In real-world recommendation scenarios, users typically engage with platforms through multiple types of behavioral interactions. Multi-behavior recommendation algorithms aim to leverage various auxiliary user behaviors to enhance prediction for target behaviors of primary interest (e.g., buy), thereby overcoming performance limitations caused by data sparsity in target behavior records. Current state-of-the-art approaches typically employ hierarchical design following either cascading (e.g., view$\rightarrow$cart$\rightarrow$buy) or parallel (unified$\rightarrow$behavior$\rightarrow$specific components) paradigms, to capture behavioral relationships. However, these methods still face two critical challenges: (1) severe distribution disparities across behaviors, and (2) negative transfer effects caused by noise in auxiliary behaviors. In this paper, we propose a novel model-agnostic Hierarchical Graph Information Bottleneck (HGIB) framework for multi-behavior recommendation to effectively address these challenges. Following information bottleneck principles, our framework optimizes the learning of compact yet sufficient representations that preserve essential information for target behavior prediction while eliminating task-irrelevant redundancies. To further mitigate interaction noise, we introduce a Graph Refinement Encoder (GRE) that dynamically prunes redundant edges through learnable edge dropout mechanisms. We conduct comprehensive experiments on three real-world public datasets, which demonstrate the superior effectiveness of our framework. Beyond these widely used datasets in the academic community, we further expand our evaluation on several real industrial scenarios and conduct an online A/B testing, showing again a significant improvement in multi-behavior recommendations. The source code of our proposed HGIB is available at https://github.com/zhy99426/HGIB.

In the dynamic environment of multimedia-sharing platforms like X (formerly known as Twitter) and TikTok, multimedia recommendation systems have been widely used to help users discover items of interest. However, traditional approaches often fall short, when the item modalities are incomplete, a common issue in real-world scenarios. To this end, we introduce the unified heterogeneous Hypergraph construction for the Incomplete multimedia REcommendation ( HIRE ), a novel framework designed to jointly learn a heterogeneous hypergraph and perform accurate recommendations under incomplete scenarios. HIRE first initializes the unified heterogeneous hypergraph for modality completion and employs self-supervised learning aligned with the contrastive text-centered view for multimedia recommendation. Such integration effectively handles the challenges posed by incomplete modalities, leading to improved recommendation accuracy. Furthermore, we find that the hypergraph directly learned from the HIRE is a dense structure which can be inaccurate and coarse. Therefore, we devise the HIRE framework with Sparse constraint named HIRES , which uniquely integrates optimal transport and a \(\ell_{2,1}\) -norm to refine the hypergraph structure. Our extensive experiments across various datasets demonstrate the superiority of HIRES in addressing incomplete modalities, establishing it as a powerful tool for personalized multimedia recommendations.

The sparse interactions between users and items have aggravated the difficulty of their representations in recommender systems. Existing methods leverage tags to alleviate the sparsity problem but ignore prevalent logical relations among items and tags (e.g., membership, hierarchy, and exclusion), which can be leveraged to enhance the accuracy of modeling user preferences and conducting recommendations. To this end, we propose to extract logical relations among item tags from existing tag taxonomies and exploit the individual strengths of the Poincaré and the Lorentz models in hyperbolic space for logical relation modeling towards enhanced recommendations. Moreover, we find that the logical relations directly extracted from existing tag taxonomies can be inaccurate and coarse. Therefore, we further devise innovative consistency-based and granularity- based weighting mechanisms based on user behavior patterns for data-driven logical relation mining that can be jointly optimized along with recommendations in an end-to-end fashion. Extensive experiments on four real-world benchmark datasets show drastic performance gains brought by our proposed framework, which constantly achieves an average of 8.25% improvement over state-of-the-art competitors regarding both Recall and NDCG metrics. Insightful case studies further demonstrate that our automatically refined logical relations are highly accurate and interpretable.

Estimating individual treatment effects (ITE) from observational data is challenging due to the absence of counterfactuals and the treatment selection bias. Prevalent ITE estimation methods tackle these challenges by aligning the treated and controlled distributions in the representational space. However, two critical issues have long been overlooked: (1)Mini-batch sampling sensitivity (MSS) issue, where representation distribution alignment at a mini-batch level is vulnerable to poor sampling cases, such as data imbalance and outliers; (2)Inconsistent representation learning (IRL) issue, where representation learning within a unified backbone network suffers from inconsistent gradient update directions due to the distribution skew between different treatment groups. To resolve these issues, we propose CE-RCFR, a Robust CounterFactual Regression framework for Consensus-Enabled causal effect estimation, including a relaxed distribution discrepancy regularizer (RDDR) module and a consensus-enabled aggregator (CEA) module. Specifically, for the robust representation alignment perspective, RDDR addresses the MSS issue by minimizing unbalanced optimal transport divergence between different treatment groups with a relaxed marginal constraint. For the accurate representation optimization perspective, CEA addresses the IRL issue by resolving the consistent gradient update directions on shared parameters within the backbone network. Extensive experiments demonstrate that CE-RCFR significantly outperforms the state-of-the-art methods in treatment effect estimations.

User cold-start recommendation aims to provide accurate items for the newly joint users and is a hot and challenging problem. Nowadays as people participant in different domains, how to recommend items in the new domain for users in an old domain has become more urgent. In this paper, we focus on the Dual Cold-Start Cross Domain Recommendation (Dual-CSCDR) problem. That is, providing the most relevant items for new users on the source and target domains. The prime task in Dual-CSCDR is to properly model user-item rating interactions and map user expressive embeddings across domains. However, previous approaches cannot solve Dual-CSCDR well, since they separate the collaborative filtering and distribution mapping process, leading to the error superimposition issue. Moreover, most of these methods fail to fully exploit the cross-domain relationship among large number of non-overlapped users, which strongly limits their performance. To fill this gap, we propose User Distribution Mapping model with Collaborative Filtering (UDMCF), a novel end-to-end cold-start cross-domain recommendation framework for the Dual-CSCDR problem. UDMCF includes two main modules, i.e., rating prediction module and distribution alignment module. The former module adopts one-hot ID vectors and multi-hot historical ratings for collaborative filtering via a contrastive loss. The latter module contains overlapped user embedding alignment and general user subgroup distribution alignment. Specifically, we innovatively propose unbalance distribution optimal transport with typical subgroup discovering algorithm to map the whole user distributions. Our empirical study on several datasets demonstrates that UDMCF significantly outperforms the state-of-the-art models under the Dual-CSCDR setting.

Cross-Domain Recommendation has been popularly studied to resolve data sparsity problem via leveraging knowledge transfer across different domains. In this paper, we focus on theUnified Cross-Domain Recommendation(Unified CDR) problem. That is, how to enhance the recommendation performance within and cross domains when users are partially overlapped. It has two main challenges, i.e., 1) how to obtain robust matching solution among the whole users and 2) how to exploit consistent and accurate results across domains. To address these two challenges, we proposeMUCRP, a cross-domain recommendation framework for the Unified CDR problem.MUCRPcontains three modules, i.e., variational rating reconstruction module, robust variational embedding alignment module, and cycle-consistent preference extraction module. To solve the first challenge, we propose fused Gromov-Wasserstein distribution co-clustering optimal transport to obtain more robust matching solution via considering both semantic and structure information. To tackle the second challenge, we propose embedding-consistent and prediction-consistent losses via dual autoencoder framework to achieve consistent results. Our empirical study on Douban and Amazon datasets demonstrates thatMUCRPsignificantly outperforms the state-of-the-art models.

Personalized diagnosis prediction based on electronic health records (EHR) of patients is a promising yet challenging task for AI in healthcare. Existing studies typically ignore the heterogeneity of diseases across different patients. For example, diabetes can have different complications across different patients (e.g., hyperlipidemia and circulatory disorder), which requires personalized diagnoses and treatments. Specifically, existing models fail to consider 1) varying severity of the same diseases for different patients, 2) complex interactions among syndromic diseases, and 3) dynamic progression of chronic diseases. In this work, we propose to perform personalized diagnosis prediction based on EHR data via capturing disease severity, interaction, and progression. In particular, we enable personalized disease representations via severity-driven embeddings at the disease level. Then, at the visit level, we propose to capture higher-order interactions among diseases that can collectively affect patients’ health status via hypergraph-based aggregation; at the patient level, we devise a personalized generative model based on neural ordinary differential equations to capture the continuous-time disease progressions underlying discrete and incomplete visits. Extensive experiments on two real-world EHR datasets show significant performance gains brought by our approach, yielding average improvements of 10.70% for diagnosis prediction over state-of-the-art competitors.

Sequential Cross-Domain Recommendation (CDR) has been popularly studied to utilize different domain knowledge and users’ historical behaviors for the next-item prediction. In this paper, we focus on the cross-domain sequential recommendation problem. This commonly exist problem is rather challenging from two perspectives, i.e., the implicit user historical rating sequences are difficult in modeling and the users/items on different domains are mostly non-overlapped. Most previous sequential CDR approaches cannot solve the cross-domain sequential recommendation problem well, since (1) they cannot sufficiently depict the users’ actual preferences, (2) they cannot leverage and transfer useful knowledge across domains. To tackle the above issues, we propose joint Internal multi-interest exploration and External domain alignment for cross domain Sequential Recommendation model (IESRec). IESRec includes two main modules, i.e., internal multi-interest exploration module and external domain alignment module. To reflect the users’ diverse characteristics with multi-interests evolution, we first propose internal temporal optimal transport method in the internal multi-interest exploration module. We further propose external alignment optimal transport method in the external domain alignment module to reduce domain discrepancy for the item embeddings. Our empirical studies on Amazon datasets demonstrate that IESRec significantly outperforms the state-of-the-art models.

The sparse interactions between users and items on the web have aggravated the difficulty of their representations in recommender systems. Existing approaches leverage item attributes (e.g., item categories and tags) to alleviate the data sparsity problem, so as to enhance the performance and interpretability of recommendation. However, directly using all attributes of items cannot avoid the negative impacts of low-quality attributes, where manually labeling the quality of attributes is time-consuming. To this end, we propose HQRec to jointly measure the quality of attributes automatically and perform recommendation accurately. Specifically, we first analyze the different qualities among item attributes, and propose to leverage item categories to select high-quality tags via category-guided quality measurement and direction-aware optimization in an unsupervised fashion. Then, we propose to capture the complex relations among users and items based on the high-quality attributes, where a novel quality-aware embedding fusion and quality-aware embedding propagation mechanism for users and items is devised. Extensive experiments on four real-world benchmark datasets show drastic performance gains brought by our proposed HQRec framework, which constantly achieves an average of 14.73% improvement over the state-of-the-art baselines in terms of Recall and NDCG metrics. Insightful case studies also show that our automatic quality measurements are highly accurate and interpretable.

The sparse interactions between users and items on the web have aggravated the difficulty of their representations in recommender systems. Existing approaches leverage tags to alleviate the data sparsity problem, so as to enhance the performance and interpretability of recommendation. However, directly using flat item tags fails to fully exploit the hierarchical relations in data, but tag taxonomies are not always available. To this end, we propose TaxoRec to jointly construct a tag taxonomy automatically and perform recommendation accurately in hyperbolic space. Specifically, we first leverage hyperbolic space and enable the optimization of a discrete taxonomy structure via a representation-aware scoring function and an adaptive clustering algorithm, and preserve the hierarchical structure for interpretability. Then, we propose to capture the complex relations among users, items, and tags in a unified hyperbolic metric space, where a novel tag-enhanced aggregation mechanism and tag-enhanced metric learning algorithm for users and items are defined. Extensive experiments on four real-world benchmark datasets show drastic performance gains brought by our proposed TaxoRec framework, which constantly achieves an average of 7.76% improvement over the state-of-the-art baselines regarding both Recall and NDCG metrics. Insightful case studies also show that our automatically constructed tag taxonomies are highly accurate and interpretable.

Drug recommendation is an important task of AI for healthcare. To recommend proper drugs, existing methods rely on various clinical records (e.g., diagnosis and procedures), which are commonly found in data such as electronic health records (EHRs). However, detailed records as such are often not available and the inputs might merely include a set of symptoms provided by doctors. Moreover, existing drug recommender systems usually treat drugs as individual items, ignoring the unique requirements that drug recommendation has to be done on a set of items (drugs), which should be as small as possible and safe without harmful drug-drug interactions (DDIs).

Cross-Domain Recommendation (CDR) has been popularly studied to utilize different domain knowledge to solve the cold-start problem in recommender systems. Most of the existing CDR models assume that both the source and target domains share the same overlapped user set for knowledge transfer. However, only few proportion of users simultaneously activate on both the source and target domains in practical CDR tasks. In this paper, we focus on the Partially Overlapped Cross-Domain Recommendation (POCDR) problem, that is, how to leverage the information of both the overlapped and non-overlapped users to improve recommendation performance. Existing approaches cannot fully utilize the useful knowledge behind the non-overlapped users across domains, which limits the model performance when the majority of users turn out to be non-overlapped. To address this issue, we propose an end-to-end Dual-autoencoder with Variational Domain-invariant Embedding Alignment (VDEA) model, a cross-domain recommendation framework for the POCDR problem, which utilizes dual variational autoencoders with both local and global embedding alignment for exploiting domain-invariant user embedding. VDEA first adopts variational inference to capture collaborative user preferences, and then utilizes Gromov-Wasserstein distribution co-clustering optimal transport to cluster the users with similar rating interaction behaviors. Our empirical studies on Douban and Amazon datasets demonstrate that VDEA significantly outperforms the state-of-the-art models, especially under the POCDR setting.

Cold-start diagnosis prediction is a challenging task for AI in healthcare, where often only a few visits per patient and a few observations per disease can be exploited. Although meta-learning is widely adopted to address the data sparsity problem in general domains, directly applying it to healthcare data is less effective, since it is unclear how to capture both the temporal relations in clinical visits and the complicated relations among syndromic diseases for precise personalized diagnosis. To this end, we first propose a novel Meta-learning framework for cold-start diagnosis prediction in healthCare data (MetaCare). By explicitly encoding the effects of disease progress over time as a generalization prior, MetaCare dynamically predicts future diagnosis and timestamp for infrequent patients. Then, to model complicated relations among rare diseases, we propose to utilize domain knowledge of hierarchical relations among diseases, and further perform diagnosis subtyping to mine the latent syndromic relations among diseases. Finally, to tailor the generic meta-learning framework with personalized parameters, we design a hierarchical patient subtyping mechanism and bridge the modeling of both infrequent patients and rare diseases. We term the joint model as MetaCare++. Extensive experiments on two real-world benchmark datasets show significant performance gains brought by MetaCare++, yielding average improvements of 7.71% for diagnosis prediction and 13.94% for diagnosis time prediction over the state-of-the-art baselines.

Implicit feedback is widely explored by modern recommender systems. Since the feedback is often sparse and imbalanced, it poses great challenges to the learning of complex interactions among users and items. Metric learning has been proposed to capture user-item interactions from implicit feedback, but existing methods only represent users and items in a single metric space, ignoring the fact that users can have multiple preferences and items can have multiple properties, which leads to potential conflicts limiting their performance in recommendation. To capture the multiple facets of user preferences and item properties while resolving their potential conflicts, we propose the novel framework of Multi-fAcet Recommender networks with Spherical optimization (MARS). By designing a cross-facet similarity measurement, we project users and items into multiple metric spaces for fine-grained representation learning, and compare them only in the proper spaces. Furthermore, we devise a spherical optimization strategy to enhance the effectiveness and robustness of the multi-facet recommendation framework. Extensive experiments on six real-world benchmark datasets show drastic performance gains brought by MARS, which constantly achieves up to 40% improvements over the state-of-the-art baselines regarding both HR and nDCG metrics.