Graph pooling has gained significant progress in recent years as an effective solution for graph-level property classification tasks. With the emergence of research on Heterogeneous Information Networks (HINs), this paper argues that graph-level datasets for graph classification should be treated as HINs rather than homogeneous graphs to enhance information aggregation. We propose HINPool, a novel and general graph pooling framework for graph-level property classification with HINs. First, we devise a systematic HIN construction procedure from the original data to capture complex interactions. Next, we introduce a type-aware heterogeneous graph pooling method featuring a Type-Aware Selector (TAS) to select essential nodes and a Readout Aggregator (RA) to fuse critical information into a graph-level representation. Finally, a cross-layer fusion function is applied to combine the output embeddings from each graph pooling layer, creating a final graph representation for downstream classification tasks. Our approach achieves near state-of-the-art performance on widely used graph classification benchmark datasets, demonstrating significant improvements in four out of five datasets. This work redefines the strategy for graph-level property classification with HGNNs and heterogeneous graph pooling to model intricate relationships, enhancing performance without requiring extensive domain-specific knowledge.

As online ecosystems grow increasingly complex, personalized recommendation systems must integrate user preferences across heterogeneous content sources and interaction scenarios. However, conventional methods typically model each source and scenario in isolation, hindering their ability to capture shared and complementary signals across contexts. In this work, we propose MM4Rec, a unified framework for multi-source and multi-scenario recommendation. MM4Rec introduces a Source-Aware Transformer Encoder to jointly model heterogeneous inputs, a Multi-Scenario Behavior Extraction Layer based on a multi-mixture-of-experts architecture to capture scenario-specific dynamics, and a Trend-Aware Learner to enhance temporal representation learning. Extensive experiments on three real-world datasets demonstrate that MM4Rec consistently outperforms strong baselines across standard recommendation metrics. To facilitate future research, we also release two large-scale datasets encompassing diverse sources and scenarios.

Sequential recommendation in e-commerce utilizes users' anonymous browsing histories to personalize product suggestions without relying on private information. Existing item ID-based methods and multimodal models often overlook the temporal alignment of modalities like textual descriptions, visual content, and prices in user browsing sequences. To address this limitation, this paper proposes the Multimodal Time-aligned Shared Token Recommender (MTSTRec), a transformer-based framework with a single time-aligned shared token per product for efficient cross-modality fusion. MTSTRec preserves the distinct contributions of each modality while aligning them temporally to better capture user preferences. Extensive experiments demonstrate that MTSTRec achieves state-of-the-art performance across multiple sequential recommendation benchmarks, significantly improving upon existing multimodal fusion. Our code is available at https://github.com/idssplab/MTSTRec.

Tags play a critical role in enhancing product discoverability, optimizing search results, and enriching recommendation systems on e-commerce platforms. Despite the recent advancements in large language models (LLMs), which have shown proficiency in processing and understanding textual information, their application in tag generation remains an under-explored yet complex challenge. To this end, we introduce a novel method for automatic product tagging using LLMs to create behavior-enhanced tags (BETags). Specifically, our approach begins by generating base tags using an LLM. These base tags are then refined into BETags by incorporating user behavior data. This method aligns the tags with users' actual browsing and purchasing behavior, enhancing the accuracy and relevance of tags to user preferences. By personalizing the base tags with user behavior data, BETags are able to capture deeper behavioral insights, which is essential for understanding nuanced user interests and preferences in e-commerce environments. Moreover, since BETags are generated offline, they do not impose real-time computational overhead and can be seamlessly integrated into downstream tasks commonly associated with recommendation systems and search optimization. Our evaluation of BETag across three datasets--- Amazon (Scientific), MovieLens-1M, and FreshFood---shows that our approach significantly outperforms both human-annotated tags and other automated methods. These results highlight BETag as a scalable and efficient solution for personalized automated tagging, advancing e-commerce platforms by creating more tailored and engaging user experiences.

Graph Neural Networks (GNNs) frequently face class imbalance issues, especially in heterogeneous graphs. Existing GNNs often assume balanced class sizes, which isn’t true in many cases. Applying them directly to imbalanced data can lead to sub-optimal performance. Traditional oversampling methods, while effective, risk overfitting and face difficulties in reintegrating synthetic samples into the original graph. In this study, we introduce Framework of Imbalanced Node Classification on heterogeneous graph neural network with GAN (FincGAN), a new framework that utilizes oversampling techniques to address class imbalance in heterogeneous graphs. Instead of duplicating existing samples, FincGAN employs a Generative Adversarial Network (GAN) to create synthetic samples and uses deep learning-based edge generators to connect them back to the original graph. Our evaluations on spam user detection in the Amazon and Yelp Review datasets show that FincGAN outperforms baseline models in all essential metrics, including F-score and AUC-PRC score, showing its effectiveness in addressing class imbalance.

Push notifications efficiently deliver real-time messages, boosting user engagement and website traffic. However, users often passively receive notifications without active interaction in recommendation contexts. Consequently, for precise recommendations, Click-Through Rate (CTR) prediction for push notifications requires addressing challenges such as user temporal and contextual preferences, the dynamic nature of user click behavior, and limited interactions between users and items. We propose Push4Rec, a novel push notification recommendation model designed explicitly for news articles. Push4Rec integrates pivotal learners to extract information adeptly. It assesses click behavior, captures preferences, and comprehends trends’ influence. A fusion function and gating network ensure versatile extraction of user click preferences. We assessed Push4Rec using a real-world push notification dataset from our partnering company. Push4Rec outperformed benchmark models, delivering state-of-the-art results across all evaluation metrics. Thus, we believe that Push4Rec, with its novel approach, sets a new standard in push notification services, driving forward the field of personalized recommendation systems.

Graph neural networks are a promising deep learning method that can apply graph structures to various tasks. In real-world scenarios, we often have heterogeneous graphs, wherein different node and edge types capture complex interactions between nodes. High-dimensional node features provide rich information about the target nodes. Conventional heterogeneous graph neural networks (HGNN) focus more on graph structures than node features and may have difficulties extracting knowledge from complex node features. In this study, we propose a novel framework, the tree-boosted heterogeneous graph neural network abbreviated as TreeXGNN, which could efficiently and automatically extract target node features via gradient-boosted decision trees (GBDT). It integrates community structure information with proper fusion modules and a shared feature space design on HGNN. We achieved state-of-the-art performance on the three well-known heterogeneous graph benchmark datasets, IMDB, DBLP, and ACM, and significantly improved performance compared to previous studies. Our work paves the foundation for integrating tree-based models to boost HGNNs for general community analysis.

BACKGROUND: Genome-wide association studies (GWAS) provide a powerful means to identify associations between genetic variants and phenotypes. However, GWAS techniques for detecting epistasis, the interactions between genetic variants associated with phenotypes, are still limited. We believe that developing an efficient and effective GWAS method to detect epistasis will be a key for discovering sophisticated pathogenesis, which is especially important for complex diseases such as Alzheimer's disease (AD). RESULTS: In this regard, this study presents GenEpi, a computational package to uncover epistasis associated with phenotypes by the proposed machine learning approach. GenEpi identifies both within-gene and cross-gene epistasis through a two-stage modeling workflow. In both stages, GenEpi adopts two-element combinatorial encoding when producing features and constructs the prediction models by L1-regularized regression with stability selection. The simulated data showed that GenEpi outperforms other widely-used methods on detecting the ground-truth epistasis. As real data is concerned, this study uses AD as an example to reveal the capability of GenEpi in finding disease-related variants and variant interactions that show both biological meanings and predictive power. CONCLUSIONS: The results on simulation data and AD demonstrated that GenEpi has the ability to detect the epistasis associated with phenotypes effectively and efficiently. The released package can be generalized to largely facilitate the studies of many complex diseases in the near future.