In this paper, we systemically disentangle BHG, a graph deep learning framework for daily users' ads recommendations. BHG mainly relies on two pillars: (1) graph tokenization to convert the input temporal heterogeneous graph into sequences of tokens, and (2) graph MLP-Mixer neural architecture to learn node representations on sequences of tokens via a mini-batch manner. In general, BHG embraces three advantages: (1) flexibility, i.e., BHG can be seamlessly integrated with any existing industrial recommendation model by treating the learned node embeddings as additional features that encode interactions, (2) efficiency, i.e., the graph tokenization allows sampling the neighborhood both locally and globally, and reduces the number of nodes considered for aggregations, and (3) model simplicity, i.e., the graph MLP-Mixer does not require self-attention for aggregating nodes and hence enjoys the simplicity. We demonstrate the superior performance of the proposed BHG on two internal datasets and one public dataset. We hope this paper can share insights and explain large-scale graph deep learning deployments for researchers, engineers, and practitioners.

Transformers serve as the backbone architectures of Foundational Models, where domain-specific tokenizers allow them to adapt to various domains. Graph Transformers (GTs) have recently emerged as leading models in geometric deep learning, outperforming Graph Neural Networks (GNNs) in various graph learning tasks. However, the development of tokenizers for graphs has lagged behind other modalities, with existing approaches relying on heuristics or GNNs co-trained with Transformers. To address this, we introduce GQT (\textbf{G}raph \textbf{Q}uantized \textbf{T}okenizer), which decouples tokenizer training from Transformer training by leveraging multi-task graph self-supervised learning, yielding robust and generalizable graph tokens. Furthermore, the GQT utilizes Residual Vector Quantization (RVQ) to learn hierarchical discrete tokens, resulting in significantly reduced memory requirements and improved generalization capabilities. By combining the GQT with token modulation, a Transformer encoder achieves state-of-the-art performance on 20 out of 22 benchmarks, including large-scale homophilic and heterophilic datasets. The implementation is publicly available at \href{https://github.com/limei0307/GQT}{https://github.com/limei0307/GQT}.

Semantic ID-based recommendation models tokenize each item into a small number of discrete tokens that preserve specific semantics, leading to better performance, scalability, and memory efficiency. While recent models adopt a generative approach, they often suffer from inefficient inference due to the reliance on resource-intensive beam search and multiple forward passes through the neural sequence model. As a result, the length of semantic IDs is typically restricted (e.g., to just 4 tokens), limiting their expressiveness. To address these challenges, we propose RPG, a lightweight framework for semantic ID-based recommendation. The key idea is to produce unordered, long semantic IDs, allowing the model to predict all tokens in parallel. We train the model to predict each token independently using a multi-token prediction loss, directly integrating semantics into the learning objective. During inference, we construct a graph connecting similar semantic IDs and guide decoding to avoid generating invalid IDs. Experiments show that scaling up semantic ID length to 64 enables RPG to outperform generative baselines by an average of 12.6% on the NDCG@10, while also improving inference efficiency. Code is available at: https://github.com/facebookresearch/RPG_KDD2025.

We study the problem of few-shot graph classification across domains with nonequivalent feature spaces by introducing three new cross-domain benchmarks constructed from publicly available datasets. We also propose an attention-based graph encoder that uses three congruent views of graphs, one contextual and two topological views, to learn representations of task-specific information for fast adaptation, and task-agnostic information for knowledge transfer. We run exhaustive experiments to evaluate the performance of contrastive and meta-learning strategies. We show that when coupled with metric-based meta-learning frameworks, the proposed encoder achieves the best average meta-test classification accuracy across all benchmarks.

A wide range of models have been proposed for Graph Generative Models, necessitating effective methods to evaluate their quality. So far, most techniques use either traditional metrics based on subgraph counting, or the representations of randomly initialized Graph Neural Networks (GNNs). We propose using representations from constrastively trained GNNs, rather than random GNNs, and show this gives more reliable evaluation metrics. Neither traditional approaches nor GNN-based approaches dominate the other, however: we give examples of graphs that each approach is unable to distinguish. We demonstrate that Graph Substructure Networks (GSNs), which in a way combine both approaches, are better at distinguishing the distances between graph datasets.

We introduce a self-supervised approach for learning node and graph level representations by contrasting structural views of graphs. We show that unlike visual representation learning, increasing the number of views to more than two or contrasting multi-scale encodings do not improve performance, and the best performance is achieved by contrasting encodings from first-order neighbors and a graph diffusion. We achieve new state-of-the-art results in self-supervised learning on 8 out of 8 node and graph classification benchmarks under the linear evaluation protocol. For example, on Cora (node) and Reddit-Binary (graph) classification benchmarks, we achieve 86.8% and 84.5% accuracy, which are 5.5% and 2.4% relative improvements over previous state-of-the-art. When compared to supervised baselines, our approach outperforms them in 4 out of 8 benchmarks.

We introduce an unsupervised multi-task model to jointly learn point and shape features on point clouds. We define three unsupervised tasks including clustering, reconstruction, and self-supervised classification to train a multi-scale graph-based encoder. We evaluate our model on shape classification and segmentation benchmarks. The results suggest that it outperforms prior state-of-the-art unsupervised models: In the ModelNet40 classification task, it achieves an accuracy of 89.1% and in ShapeNet segmentation task, it achieves an mIoU of 68.2 and accuracy of 88.6%.

Classification is a supervised learning technique that predicts the classes of unobserved data by employing a model built from available data. One of the efficient ways to represent this predictive model is to express it as an optimal set of classification rules to provide comprehensibility and precision, simultaneously. In this paper, we propose a novel incremental parallel Particle Swarm Optimization (PSO) approach for classification rule discovery. Our proposed method separates the training data into a set of data chunks regarding the classes and extracts optimal set of classification rules for each chunk in a parallel manner. In order to extract the rules from data chunks, we introduce an incremental PSO algorithm in which the previously extracted rules are directly employed to initialize the swarm population. Moreover, in each generation of the swarm, a tournament method is employed to substitute the weak individuals with strong extracted knowledge. To support the parallelism, we assign a PSO thread for each data chunk. As soon as all the PSO threads are completed, the extracted rules are integrated into a rule-base to construct a classification model. The evaluation results of the proposed approach on six datasets suggest that the classification precision of our proposed framework is competitive with offline learning methods and is 35% faster than its counterpart offline PSO approach.