Generative recommendation (GR) has shown great promise in industrial applications, particularly for candidate generation and end-to-end recommendations. However, existing GR training paradigms suffer from two fundamental mismatches with real-world deployment requirements. First, they optimize for point-wise prediction of a single ground-truth item, whereas practical systems must produce a diverse, high-value set of candidates. Second, they treat all user interactions as equally informative, ignoring their inherent differences in utility. Although reward-based fine-tuning offers a partial remedy, it often lacks token-level supervision. To address these challenges, we reformulate GR as a sequential set-generation problem and propose GFlowGR, a GFlowNet-based fine-tuning framework that explicitly aligns generation probabilities with item-level utilities. GFlowGR comprises three tightly integrated components, each addressing a key limitation of conventional fine-tuning: a trajectory sampler that constructs training trajectories from candidate sets to enable set-wise learning, a behavior-aware reward model that quantifies item utility to support value-aware optimization, and a GFlowNet objective that provides token-level supervision. Extensive experiments on three real-world datasets with two representative LLM-based GR backbones show consistent and significant improvements over strong baselines, validating the effectiveness of our approach. For real-world deployment, GFlowGR has been integrated into Taobao 's search advertising businesses, delivering a 0.4% relative improvement in annual revenue since its launch in mid-2025, corresponding to billion-level monetary gains. Code is available at https://github.com/Applied-Machine-Learning-Lab/SIGIR26_GFlowGR.

Sequential recommendation (SR) aims to capture users' dynamic interests and sequential patterns based on their historical interactions. Recently, the powerful capabilities of large language models (LLMs) have driven their adoption in SR. However, we identify two critical challenges in existing LLM-based SR methods: 1) embedding collapse when incorporating pre-trained collaborative embeddings and 2) catastrophic forgetting of quantized embeddings when utilizing semantic IDs. These issues dampen the model scalability and lead to suboptimal recommendation performance. Therefore, based on LLMs like Llama3-8B-instruct, we introduce a novel SR framework named MME-SID, which integrates multimodal embeddings and quantized embeddings to mitigate embedding collapse. Additionally, we propose a Multimodal Residual Quantized Variational Autoencoder (MM-RQ-VAE) with maximum mean discrepancy as the reconstruction loss and contrastive learning for alignment, which effectively preserve intra-modal distance information and capture inter-modal correlations, respectively. To further alleviate catastrophic forgetting, we initialize the model with the trained multimodal code embeddings. Finally, we fine-tune the LLM efficiently using LoRA in a multimodal frequency-aware fusion manner. Extensive experiments on three public datasets validate the superior performance of MME-SID thanks to its capability to mitigate embedding collapse and catastrophic forgetting. The implementation code and datasets are publicly available for reproduction: https://github.com/Applied-Machine-Learning-Lab/MME-SID.

Site recommendation, which aims at predicting the optimal location for brands to open new branches, has demonstrated an important role in assisting decision-making in modern business. In contrast to traditional recommender systems that can benefit from extensive information, site recommendation starkly suffers from extremely limited information and thus leads to unsatisfactory performance. Therefore, existing site recommendation methods primarily focus on several specific name brands and heavily rely on fine-grained human-crafted features to avoid the data sparsity problem. However, such solutions are not able to fulfill the demand for rapid development in modern business. Therefore, we aim to alleviate the data sparsity problem by effectively utilizing data across multiple cities and thereby propose a novel Optimal Transport enhanced Cross-city (OTC) framework for site recommendation. Specifically, OTC leverages optimal transport (OT) on the learned embeddings of brands and regions separately to project the brands and regions from the source city to the target city. Then, the projected embeddings of brands and regions are utilized to obtain the inference recommendation in the target city. By integrating the original recommendation and the inference recommendations from multiple cities, OTC is able to achieve enhanced recommendation results. The experimental results on the real-world OpenSiteRec dataset, encompassing thousands of brands and regions across four metropolises, demonstrate the effectiveness of our proposed OTC in further improving the performance of site recommendation models.

As a representative information retrieval task, site recommendation, which aims at predicting the optimal sites for a brand or an institution to open new branches in an automatic data-driven way, is beneficial and crucial for brand development in modern business. However, there is no publicly available dataset so far and most existing approaches are limited to an extremely small scope of brands, which seriously hinders the research on site recommendation. Therefore, we collect, construct and release an open comprehensive dataset, namely OpenSiteRec, to facilitate and promote the research on site recommendation. Specifically, OpenSiteRec leverages a heterogeneous graph schema to represent various types of real-world entities and relations in four international metropolises. To evaluate the performance of the existing general methods on the site recommendation task, we conduct benchmarking experiments of several representative recommendation models on OpenSiteRec. Furthermore, we also highlight the potential application directions to demonstrate the wide applicability of OpenSiteRec. We believe that our OpenSiteRec dataset is significant and anticipated to encourage the development of advanced methods for site recommendation. OpenSiteRec is available online at https://OpenSiteRec.github.io/.

The bursting of media sharing platforms like TikTok, YouTube, and Kwai enables normal users to create and share content with worldwide audiences. The most popular YouTuber can attract up to 100 million followers. Since there are multiple popular platforms, it’s quite common that a YouTuber publishes the same media to multiple platforms, or replicates all media from one platform to another. However, the users of different platforms have different tastes. The media that is popular on one platform may not be a great vogue on other platforms. Observing such cross-platform variance, we propose a new task: estimating the user engagement score of a media on one platform given its popularity on other platforms. This task can benefit both the YouTubers and the platform. On one hand, YouTubers can use the predicted engagement to guide the media reworking; on the other hand, the platform can use the predicted engagement to establish promotion and advertising plans. Therefore, this task is of great practical value. To tackle this task, we propose a disentangled neural network that can separate the general media adorability from platform inclinations. In this manner, by substituting the inclination from the source platform to the target platform, we are able to predict the user engagement in the target platform. To validate the proposed model, we manage to build a dataset of micro-videos which are published on four platforms TikTok, Kwai, Bilibili, and WESEE. The experimental results prove the effectiveness of the proposed model.

Accurate prediction of drug-drug interaction (DDI) is crucial to achieving effective decision-making in medical treatment for both doctors and patients. Recently, many deep learning based methods have been proposed to learn from drug-related features and conduct DDI prediction. These works have achieved promising results. However, the extreme imbalance of medical data poses a serious problem to DDI prediction, where a small fraction of DDI types occupy the majority training data. A straightforward way is to develop an appropriate policy to sample the data. Due to the high complexity and speciality of medical science, a dynamic learnable policy is required instead of a heuristic, uniform or static one. Therefore, we propose a REinforced Student-Teacher curriculum learning model (REST) for effective sampling to tackle this imbalance problem. Specifically, REST consists of two interactive parts, which are a heterogeneous graph neural network as the student and a reinforced sampler as the teacher. In each interaction, the teacher model takes action to sample an appropriate batch to train the student model according to the student model state while the cumulated improvement in performance of the student model is treated as the reward for policy gradient of the teacher model. The experimental results on two benchmarking datasets have demonstrated the significant effectiveness of our proposed model in DDI prediction, especially for the DDI types with low frequency.

The aim of ICD coding is to assign International Classification of Diseases (ICD) codes to unstructured clinical notes or discharge summaries. Numerous methods have been proposed for automatic ICD coding in an effort to reduce human labor and errors. However, existing works disregard the data imbalance problem of clinical notes. In addition, the noisy clinical note issue has not been thoroughly investigated. To address such issues, we propose a knowledge enhanced Graph Attention Network (GAT) under multi-task learning setting. Specifically, multi-level information transitions and interactions have been implemented. On the one hand, a large heterogeneous text graph is constructed to capture both intra- and inter-note correlations between various semantic concepts, thereby alleviating the data imbalance issue. On the other hand, two auxiliary healthcare tasks have been proposed to facilitate the sharing of information across tasks. Moreover, to tackle the issue of noisy clinical notes, we propose to utilize the rich structured knowledge facts and information provided by medical domain knowledge, thereby encouraging the model to focus on the clinical notes' noteworthy portion and valuable information. The experimental results on the widely-used medical dataset, MIMIC-III, demonstrate the advantages of our proposed framework.

Link prediction aims to identify potential missing triples in knowledge graphs. To get better results, some recent studies have introduced multimodal information to link prediction. However, these methods utilize multimodal information separately and neglect the complicated interaction between different modalities. In this paper, we aim at better modeling the inter-modality information and thus introduce a novel Interactive Multimodal Fusion (IMF) model to integrate knowledge from different modalities. To this end, we propose a two-stage multimodal fusion framework to preserve modality-specific knowledge as well as take advantage of the complementarity between different modalities. Instead of directly projecting different modalities into a unified space, our multimodal fusion module limits the representations of different modalities independent while leverages bilinear pooling for fusion and incorporates contrastive learning as additional constraints. Furthermore, the decision fusion module delivers the learned weighted average over the predictions of all modalities to better incorporate the complementarity of different modalities. Our approach has been demonstrated to be effective through empirical evaluations on several real-world datasets. The implementation code is available online at https://github.com/HestiaSky/IMF-Pytorch.

Cross-Domain Recommendation (CDR) has attracted increasing attention in recent years as a solution to the data sparsity issue. The fundamental paradigm of prior efforts is to train a mapping function based on the overlapping users/items and then apply it to the knowledge transfer. However, due to the commercial privacy policy and the sensitivity of user data, it is unrealistic to explicitly share the user mapping relations and behavior data. Therefore, in this paper, we consider a more practical cross-domain scenario, where there is no explicit overlap between the source and target domains in terms of users/items. Since the user sets of both domains are drawn from the entire population, there may be commonalities between their user characteristics, resulting in comparable user preference distributions. Thus, without the mapping relations at user level, it is feasible to model this distribution-level relation to transfer knowledge between domains. To this end, we propose a novel framework that improves the effect of representation learning on the target domain by aligning the representation distributions between the source and target domains. In addition, GWCDR can be easily integrated with existing single-domain collaborative filtering methods to achieve cross-domain recommendation. Extensive experiments on two pairs of public bidirectional datasets demonstrate the effectiveness of our proposed framework in enhancing the recommendation performance.