Kangfei Zhao

Chaohao Yuan, Zhenjie Song, Ercan Engin Kuruoglu et al.

Graph Transformers (GTs) have emerged as a promising graph learning tool, leveraging their all-pair connected property to effectively capture global information. To address the over-smoothing problem in deep GNNs, global attention was initially introduced, eliminating the necessity for using deep GNNs. However, through empirical and theoretical analysis, we verify that the introduced global attention exhibits severe over-smoothing, causing node representations to become indistinguishable due to its inherent low-pass filtering. This effect is even stronger than that observed in GNNs. To mitigate this, we propose PageRank Transformer (ParaFormer), which features a PageRank-enhanced attention module designed to mimic the behavior of deep Transformers. We theoretically and empirically demonstrate that ParaFormer mitigates over-smoothing by functioning as an adaptive-pass filter. Experiments show that ParaFormer achieves consistent performance improvements across both node classification and graph classification tasks on 11 datasets ranging from thousands to millions of nodes, validating its efficacy. The supplementary material, including code and appendix, can be found in https://github.com/chaohaoyuan/ParaFormer.

Jianheng Tang, Weiqi Zhang, Jiajin Li et al.

Graph alignment, which aims at identifying corresponding entities across multiple networks, has been widely applied in various domains. As the graphs to be aligned are usually constructed from different sources, the inconsistency issues of structures and features between two graphs are ubiquitous in real-world applications. Most existing methods follow the ``embed-then-cross-compare'' paradigm, which computes node embeddings in each graph and then processes node correspondences based on cross-graph embedding comparison. However, we find these methods are unstable and sub-optimal when structure or feature inconsistency appears. To this end, we propose SLOTAlign, an unsupervised graph alignment framework that jointly performs Structure Learning and Optimal Transport Alignment. We convert graph alignment to an optimal transport problem between two intra-graph matrices without the requirement of cross-graph comparison. We further incorporate multi-view structure learning to enhance graph representation power and reduce the effect of structure and feature inconsistency inherited across graphs. Moreover, an alternating scheme based algorithm has been developed to address the joint optimization problem in SLOTAlign, and the provable convergence result is also established. Finally, we conduct extensive experiments on six unsupervised graph alignment datasets and the DBP15K knowledge graph (KG) alignment benchmark dataset. The proposed SLOTAlign shows superior performance and strongest robustness over seven unsupervised graph alignment methods and five specialized KG alignment methods.

Haotian Xu, Kangfei Zhao, Jiadong Xie

Large Language Model (LLM) serving systems must balance task performance against monetary cost. Two prominent optimization techniques have emerged independently: LLM routing, which directs each query to the most cost-effective model in a model pool, and batch prompting, which packs multiple queries into a single invocation to amortize the fixed cost of the shared system prompt. These two techniques are logically complementary; i.e., routing optimizes the model assignment dimension while batching optimizes the query aggregation dimension, jointly reshaping the landscape of model utility and monetary cost. However, existing approaches explore only one side of this decision space. On the basis of empirical studies on their impacts, we are motivated to jointly optimize these two dimensions in this paper. We formulate the Route with Batching Problem, which jointly determines the target model and batch size for each query under a total cost budget, and prove it NP-hard. To solve this challenging problem, we propose RoBatch, a unified two-stage framework. In the modeling stage, RoBatch constructs a batch-aware proxy utility model that decomposes combinatorial utility estimation into utility estimation without batching and recalibration of model-specific utility degradation with batching. In the routing stage, RoBatch employs a greedy scheduling algorithm that progressively upgrades the assignment of the target model and batch size for queries along the cost-utility Pareto frontier until the budget is exhausted. Extensive experiments on six benchmarks across two LLM families (Qwen3 and Gemma3) demonstrate that RoBatch consistently achieves a superior cost-performance Pareto frontier compared with LLM routing and batch prompting baselines.

Kangkang Qi, Dongyang Xie, Wenbo Li et al.

The integration of Large Language Models (LLMs) into data analytics has unlocked powerful capabilities for reasoning over bulk structured and unstructured data. However, existing systems typically rely on either DataFrame primitives, which lack the efficient execution infrastructure of modern DBMSs, or SQL User-Defined Functions (UDFs), which isolate semantic logic from the query optimizer and burden users with implementation complexities. The LLM-powered semantic operators also bring new challenges due to the high cost and non-deterministic nature of LLM invocation, where conventional optimization rules and cost models are inapplicable for their optimization. To bridge these gaps, we present Sema, a high-performance semantic query engine built on DuckDB that treats LLM-powered semantic operators as first-class citizens. Sema introduces SemaSQL, a declarative dialect that allows users seamlessly inject natural language expressions into standard SQL clauses, enabling end-to-end optimization and execution. At the logical level, the optimizer of Sema compresses natural language expressions and deduces relational constraints from semantic operators. At runtime, Sema employs Adaptive Query Execution (AQE) to dynamically reorder operators, fuse semantic operations, and apply prompt batching. This approach seeks a Pareto-optimal execution path balancing token consumption and latency under accuracy constraints. We evaluate Sema on 20 semantic queries across classification, summarization, and extraction tasks. Experimental results demonstrate that Sema achieves $2-10 \times$ speedup against three baseline systems while achieving competitive result quality.

Yiqiang Yi, Xu Wan, Kangfei Zhao et al.

Binding affinity prediction of three-dimensional (3D) protein ligand complexes is critical for drug repositioning and virtual drug screening. Existing approaches transform a 3D protein-ligand complex to a two-dimensional (2D) graph, and then use graph neural networks (GNNs) to predict its binding affinity. However, the node and edge features of the 2D graph are extracted based on invariant local coordinate systems of the 3D complex. As a result, the method can not fully learn the global information of the complex, such as, the physical symmetry and the topological information of bonds. To address these issues, we propose a novel Equivariant Line Graph Network (ELGN) for affinity prediction of 3D protein ligand complexes. The proposed ELGN firstly adds a super node to the 3D complex, and then builds a line graph based on the 3D complex. After that, ELGN uses a new E(3)-equivariant network layer to pass the messages between nodes and edges based on the global coordinate system of the 3D complex. Experimental results on two real datasets demonstrate the effectiveness of ELGN over several state-of-the-art baselines.

Jie Tan, Yu Rong, Kangfei Zhao et al.

Combinatorial medication recommendation(CMR) is a fundamental task of healthcare, which offers opportunities for clinical physicians to provide more precise prescriptions for patients with intricate health conditions, particularly in the scenarios of long-term medical care. Previous research efforts have sought to extract meaningful information from electronic health records (EHRs) to facilitate combinatorial medication recommendations. Existing learning-based approaches further consider the chemical structures of medications, but ignore the textual medication descriptions in which the functionalities are clearly described. Furthermore, the textual knowledge derived from the EHRs of patients remains largely underutilized. To address these issues, we introduce the Natural Language-Assisted Multi-modal Medication Recommendation(NLA-MMR), a multi-modal alignment framework designed to learn knowledge from the patient view and medication view jointly. Specifically, NLA-MMR formulates CMR as an alignment problem from patient and medication modalities. In this vein, we employ pretrained language models(PLMs) to extract in-domain knowledge regarding patients and medications, serving as the foundational representation for both modalities. In the medication modality, we exploit both chemical structures and textual descriptions to create medication representations. In the patient modality, we generate the patient representations based on textual descriptions of diagnosis, procedure, and symptom. Extensive experiments conducted on three publicly accessible datasets demonstrate that NLA-MMR achieves new state-of-the-art performance, with a notable average improvement of 4.72% in Jaccard score. Our source code is publicly available on https://github.com/jtan1102/NLA-MMR_CIKM_2024.

Xiang Wu, Rong-Hua Li, Xunkai Li et al.

Graph coarsening reduces the size of a graph while preserving certain properties. Most existing methods preserve either spectral or spatial characteristics. Recent research has shown that preserving topological features helps maintain the predictive performance of graph neural networks (GNNs) trained on the coarsened graph but suffers from exponential time complexity. To address these problems, we propose Scalable Topology-Preserving Graph Coarsening (STPGC) by introducing the concepts of graph strong collapse and graph edge collapse extended from algebraic topology. STPGC comprises three new algorithms, GStrongCollapse, GEdgeCollapse, and NeighborhoodConing based on these two concepts, which eliminate dominated nodes and edges while rigorously preserving topological features. We further prove that STPGC preserves the GNN receptive field and develop approximate algorithms to accelerate GNN training. Experiments on node classification with GNNs demonstrate the efficiency and effectiveness of STPGC.

Jinhao Zhang, Kangfei Zhao, Qiuhao Zeng et al.

Transformer-based architectures have become the dominant paradigm for Continuous-Time Dynamic Graph (CTDG) learning, yet their performance remains limited on temporally shifted datasets. In this work, we identify attention dispersion as a shared failure mode of dynamic graph Transformers under temporal distribution shift. Through controlled ablation contrasting structurally and temporally distinguished historical neighbors against random ones, we show that prediction depends on a class of critical nodes that carry consistently more predictive signal than arbitrary neighbors. However, existing Transformers fail to focus on these nodes even when they are present in the input, as temporal shift weakens attention contrast and produces overly dispersed attention distributions. This diagnosis suggests a simple and transferable fix: replace standard attention with differential attention, which suppresses common-mode attention and amplifies distinctive token-level signals. When added to three representative CTDG Transformer baselines, differential attention consistently improves performance, with gains concentrated on high-shift datasets. Attention-level measurements further confirm the mechanism, showing reduced attention entropy and increased attention mass on critical nodes. Building on these findings, we introduce DiffDyG, a reference implementation combining differential attention with standard input encodings. Across 9 benchmarks and three negative sampling protocols, DiffDyG achieves SOTA performance, with especially large gains on the most shifted datasets.

Jianheng Tang, Kangfei Zhao, Jia Li

Entity alignment is the task of identifying corresponding entities across different knowledge graphs (KGs). Although recent embedding-based entity alignment methods have shown significant advancements, they still struggle to fully utilize KG structural information. In this paper, we introduce FGWEA, an unsupervised entity alignment framework that leverages the Fused Gromov-Wasserstein (FGW) distance, allowing for a comprehensive comparison of entity semantics and KG structures within a joint optimization framework. To address the computational challenges associated with optimizing FGW, we devise a three-stage progressive optimization algorithm. It starts with a basic semantic embedding matching, proceeds to approximate cross-KG structural and relational similarity matching based on iterative updates of high-confidence entity links, and ultimately culminates in a global structural comparison between KGs. We perform extensive experiments on four entity alignment datasets covering 14 distinct KGs across five languages. Without any supervision or hyper-parameter tuning, FGWEA surpasses 21 competitive baselines, including cutting-edge supervised entity alignment methods. Our code is available at https://github.com/squareRoot3/FusedGW-Entity-Alignment.

Chaohao Yuan, Kangfei Zhao, Ercan Engin Kuruoglu et al.

Graph Transformers (GTs) have demonstrated a strong capability in modeling graph structures by addressing the intrinsic limitations of graph neural networks (GNNs), such as over-smoothing and over-squashing. Recent studies have proposed diverse architectures, enhanced explainability, and practical applications for Graph Transformers. In light of these rapid developments, we conduct a comprehensive review of Graph Transformers, covering aspects such as their architectures, theoretical foundations, and applications within this survey. We categorize the architecture of Graph Transformers according to their strategies for processing structural information, including graph tokenization, positional encoding, structure-aware attention and model ensemble. Furthermore, from the theoretical perspective, we examine the expressivity of Graph Transformers in various discussed architectures and contrast them with other advanced graph learning algorithms to discover the connections. Furthermore, we provide a summary of the practical applications where Graph Transformers have been utilized, such as molecule, protein, language, vision, traffic, brain and material data. At the end of this survey, we will discuss the current challenges and prospective directions in Graph Transformers for potential future research.

Rui Li, Kangfei Zhao, Jeffrey Xu Yu et al.

Query-driven learned estimators are accurate, flexible, and lightweight alternatives to traditional estimators in query optimization. However, existing query-driven approaches struggle with the Out-of-distribution (OOD) problem, where the test workload distribution differs from the training workload, leading to performancedegradation. In this paper, we present CardOOD, a general learning framework designed to construct robust query-driven cardinality estimators that are resilient against the OOD problem. Our framework focuses on offline training algorithms that develop one-off models from a static workload, suitable for model initialization and periodic retraining. In CardOOD, we extend classical transfer/robust learning techniques to train query-driven cardinalityestimators, and the algorithms fall into three categories: representation learning, data manipulation, and new learning strategies. As these learning techniques are originally evaluated in computervision tasks, we also propose a new learning algorithm that exploits the property of cardinality estimation. This algorithm, lying in the category of new learning strategy, models the partial order constraint of cardinalities by a self-supervised learning task. Comprehensive experimental studies demonstrate the efficacy of the algorithms of CardOOD in mitigating the OOD problem to varying extents. We further integrate CardOOD into PostgreSQL, showcasing its practical utility in query optimization.

Xingyi Zhang, Kun Xie, Ningqiao Huang et al.

Recent advancements in protein design have leveraged diffusion models to generate structural scaffolds, followed by a process known as protein inverse folding, which involves sequence inference on these scaffolds. However, these methodologies face significant challenges when applied to hyper-variable structures such as antibody Complementarity-Determining Regions (CDRs), where sequence inference frequently results in non-functional sequences due to hallucinations. Distinguished from prevailing protein inverse folding approaches, this paper introduces Igseek, a novel structure-retrieval framework that infers CDR sequences by retrieving similar structures from a natural antibody database. Specifically, Igseek employs a simple yet effective multi-channel equivariant graph neural network to generate high-quality geometric representations of CDR backbone structures. Subsequently, it aligns sequences of structurally similar CDRs and utilizes structurally conserved sequence motifs to enhance inference accuracy. Our experiments demonstrate that Igseek not only proves to be highly efficient in structural retrieval but also outperforms state-of-the-art approaches in sequence recovery for both antibodies and T-Cell Receptors, offering a new retrieval-based perspective for therapeutic protein design.

Shuheng Fang, Kangfei Zhao, Yu Rong et al.

Cold-start rating prediction is a fundamental problem in recommender systems that has been extensively studied. Many methods have been proposed that exploit explicit relations among existing data, such as collaborative filtering, social recommendations and heterogeneous information network, to alleviate the data insufficiency issue for cold-start users and items. However, the explicit relations constructed based on data between different roles may be unreliable and irrelevant, which limits the performance ceiling of the specific recommendation task. Motivated by this, in this paper, we propose a flexible framework dubbed heterogeneous interaction rating network (HIRE). HIRE dose not solely rely on the pre-defined interaction pattern or the manually constructed heterogeneous information network. Instead, we devise a Heterogeneous Interaction Module (HIM) to jointly model the heterogeneous interactions and directly infer the important interactions via the observed data. In the experiments, we evaluate our model under three cold-start settings on three real-world datasets. The experimental results show that HIRE outperforms other baselines by a large margin. Furthermore, we visualize the inferred interactions of HIRE to confirm the contribution of our model.

Nan Hou, Kangfei Zhao, Jiadong Xie et al.

Large language models (LLMs) are increasingly used for semantic query processing over large corpora. A set of semantic operators derived from relational algebra has been proposed to provide a unified interface for expressing such queries, among which the semantic filter operator serves as a cornerstone. Given a table T with a natural language predicate e, for each tuple in the relation, the execution of a semantic filter proceeds by constructing an input prompt that combines the predicate e with its content, querying the LLM, and obtaining the binary decision. However, this tuple-by-tuple evaluation necessitates a complete linear scan of the table, incurring prohibitive latency and token costs. Although recent work has attempted to optimize semantic filtering, it still does not break the linear LLM invocation barriers. To address this, we propose Clustering-Sampling-Voting (CSV), a new framework that reduces LLM invocations to sublinear complexity while providing error guarantees. CSV embeds tuples into semantic clusters, samples a small subset for LLM evaluation, and infers cluster-level labels via two proposed voting strategies: UniVote, which aggregates labels uniformly, and SimVote, which weights votes by semantic similarity. Moreover, CSV triggers re-clustering on ambiguous clusters to ensure robustness across diverse datasets. The results conducted on real-world datasets demonstrate that CSV reduces the number of LLM calls by 1.28-355x compared to the state-of-the-art approaches, while maintaining comparable effectiveness in terms of Accuracy and F1 score.

Tian Ma, Kaiyu Feng, Yu Rong et al.

Personality prediction from social media posts is a critical task that implies diverse applications in psychology and sociology. The Myers Briggs Type Indicator (MBTI), a popular personality inventory, has been traditionally predicted by machine learning (ML) and deep learning (DL) techniques. Recently, the success of Large Language Models (LLMs) has revealed their huge potential in understanding and inferring personality traits from social media content. However, directly exploiting LLMs for MBTI prediction faces two key challenges: the hallucination problem inherent in LLMs and the naturally imbalanced distribution of MBTI types in the population. In this paper, we propose PostToPersonality (PtoP), a novel LLM based framework for MBTI prediction from social media posts of individuals. Specifically, PtoP leverages Retrieval Augmented Generation with in context learning to mitigate hallucination in LLMs. Furthermore, we fine tune a pretrained LLM to improve model specification in MBTI understanding with synthetic minority oversampling, which balances the class imbalance by generating synthetic samples. Experiments conducted on a real world social media dataset demonstrate that PtoP achieves state of the art performance compared with 10 ML and DL baselines.

Shuheng Fang, Kangfei Zhao, Rener Zhang et al.

In this paper, we propose PLACE (Prompt Learning for Attributed Community Search), an innovative graph prompt learning framework for ACS. Enlightened by prompt-tuning in Natural Language Processing (NLP), where learnable prompt tokens are inserted to contextualize NLP queries, PLACE integrates structural and learnable prompt tokens into the graph as a query-dependent refinement mechanism, forming a prompt-augmented graph. Within this prompt-augmented graph structure, the learned prompt tokens serve as a bridge that strengthens connections between graph nodes for the query, enabling the GNN to more effectively identify patterns of structural cohesiveness and attribute similarity related to the specific query. We employ an alternating training paradigm to optimize both the prompt parameters and the GNN jointly. Moreover, we design a divide-and-conquer strategy to enhance scalability, supporting the model to handle million-scale graphs. Extensive experiments on 9 real-world graphs demonstrate the effectiveness of PLACE for three types of ACS queries, where PLACE achieves higher F1 scores by 22% compared to the state-of-the-arts on average.

Xiang Wu, Xunkai Li, Rong-Hua Li et al.

Dynamic graphs (DGs), which capture time-evolving relationships between graph entities, have widespread real-world applications. To efficiently encode DGs for downstream tasks, most dynamic graph neural networks follow the traditional message-passing mechanism and extend it with time-based techniques. Despite their effectiveness, the growth of historical interactions introduces significant scalability issues, particularly in industry scenarios. To address this limitation, we propose ScaDyG, with the core idea of designing a time-aware scalable learning paradigm as follows: 1) Time-aware Topology Reformulation: ScaDyG first segments historical interactions into time steps (intra and inter) based on dynamic modeling, enabling weight-free and time-aware graph propagation within pre-processing. 2) Dynamic Temporal Encoding: To further achieve fine-grained graph propagation within time steps, ScaDyG integrates temporal encoding through a combination of exponential functions in a scalable manner. 3) Hypernetwork-driven Message Aggregation: After obtaining the propagated features (i.e., messages), ScaDyG utilizes hypernetwork to analyze historical dependencies, implementing node-wise representation by an adaptive temporal fusion. Extensive experiments on 12 datasets demonstrate that ScaDyG performs comparably well or even outperforms other SOTA methods in both node and link-level downstream tasks, with fewer learnable parameters and higher efficiency.

Erxue Min, Runfa Chen, Yatao Bian et al.

Recently, Transformer model, which has achieved great success in many artificial intelligence fields, has demonstrated its great potential in modeling graph-structured data. Till now, a great variety of Transformers has been proposed to adapt to the graph-structured data. However, a comprehensive literature review and systematical evaluation of these Transformer variants for graphs are still unavailable. It's imperative to sort out the existing Transformer models for graphs and systematically investigate their effectiveness on various graph tasks. In this survey, we provide a comprehensive review of various Graph Transformer models from the architectural design perspective. We first disassemble the existing models and conclude three typical ways to incorporate the graph information into the vanilla Transformer: 1) GNNs as Auxiliary Modules, 2) Improved Positional Embedding from Graphs, and 3) Improved Attention Matrix from Graphs. Furthermore, we implement the representative components in three groups and conduct a comprehensive comparison on various kinds of famous graph data benchmarks to investigate the real performance gain of each component. Our experiments confirm the benefits of current graph-specific modules on Transformer and reveal their advantages on different kinds of graph tasks.

Yuli Jiang, Yu Rong, Hong Cheng et al.

Given one or more query vertices, Community Search (CS) aims to find densely intra-connected and loosely inter-connected structures containing query vertices. Attributed Community Search (ACS), a related problem, is more challenging since it finds communities with both cohesive structures and homogeneous vertex attributes. However, most methods for the CS task rely on inflexible pre-defined structures and studies for ACS treat each attribute independently. Moreover, the most popular ACS strategies decompose ACS into two separate sub-problems, i.e., the CS task and subsequent attribute filtering task. However, in real-world graphs, the community structure and the vertex attributes are closely correlated to each other. This correlation is vital for the ACS problem. In this paper, we propose Graph Neural Network models for both CS and ACS problems, i.e., Query Driven-GNN and Attributed Query Driven-GNN. In QD-GNN, we combine the local query-dependent structure and global graph embedding. In order to extend QD-GNN to handle attributes, we model vertex attributes as a bipartite graph and capture the relation between attributes by constructing GNNs on this bipartite graph. With a Feature Fusion operator, AQD-GNN processes the structure and attribute simultaneously and predicts communities according to each attributed query. Experiments on real-world graphs with ground-truth communities demonstrate that the proposed models outperform existing CS and ACS algorithms in terms of both efficiency and effectiveness. More recently, an interactive setting for CS is proposed that allows users to adjust the predicted communities. We further verify our approaches under the interactive setting and extend to the attributed context. Our method achieves 2.37% and 6.29% improvements in F1-score than the state-of-the-art model without attributes and with attributes respectively.

Jia Li, Tomasyu Yu, Jiajin Li et al.

Graph Neural Networks (GNNs) and Variational Autoencoders (VAEs) have been widely used in modeling and generating graphs with latent factors. However, there is no clear explanation of what these latent factors are and why they perform well. In this work, we present Dirichlet Graph Variational Autoencoder (DGVAE) with graph cluster memberships as latent factors. Our study connects VAEs based graph generation and balanced graph cut, and provides a new way to understand and improve the internal mechanism of VAEs based graph generation. Specifically, we first interpret the reconstruction term of DGVAE as balanced graph cut in a principled way. Furthermore, motivated by the low pass characteristics in balanced graph cut, we propose a new variant of GNN named Heatts to encode the input graph into cluster memberships. Heatts utilizes the Taylor series for fast computation of heat kernels and has better low pass characteristics than Graph Convolutional Networks (GCN). Through experiments on graph generation and graph clustering, we demonstrate the effectiveness of our proposed framework.

Kangfei Zhao, Shengcai Liu, Yu Rong et al.

The Travelling Salesman Problem (TSP) is a classical NP-hard problem and has broad applications in many disciplines and industries. In a large scale location-based services system, users issue TSP queries concurrently, where a TSP query is a TSP instance with $n$ points. In the literature, many advanced TSP solvers are developed to find high-quality solutions. Such solvers can solve some TSP instances efficiently but may take an extremely long time for some other instances. Due to the diversity of TSP instances, it is well-known that there exists no universal best solver dominating all other solvers on all possible TSP instances. To solve TSP efficiently, in addition to developing new TSP solvers, it needs to find a per-instance solver for each TSP instance, which is known as the TSP solver selection problem. In this paper, for the first time, we propose a deep learning framework, \CTAS, for TSP solver selection in an end-to-end manner. Specifically, \CTAS exploits deep convolutional neural networks to extract informative features from TSP instances and involves data argumentation strategies to handle the scarcity of labeled TSP instances. Moreover, to support large scale TSP solver selection, we construct a challenging TSP benchmark dataset with 6,000 instances, which is known as the largest TSP benchmark. Our \CTAS achieves over 2$\times$ speedup of the average running time, comparing the single best solver, and outperforms the state-of-the-art statistical models.

Kangfei Zhao, Yu Rong, Jeffrey Xu Yu et al.

Graph representation learning has achieved a remarkable success in many graph-based applications, such as node classification, link prediction, and community detection. These models are usually designed to preserve the vertex information at different granularity and reduce the problems in discrete space to some machine learning tasks in continuous space. However, regardless of the fruitful progress, for some kind of graph applications, such as graph compression and edge partition, it is very hard to reduce them to some graph representation learning tasks. Moreover, these problems are closely related to reformulating a global layout for a specific graph, which is an important NP-hard combinatorial optimization problem: graph ordering. In this paper, we propose to attack the graph ordering problem behind such applications by a novel learning approach. Distinguished from greedy algorithms based on predefined heuristics, we propose a neural network model: Deep Order Network (DON) to capture the hidden locality structure from partial vertex order sets. Supervised by sampled partial order, DON has the ability to infer unseen combinations. Furthermore, to alleviate the combinatorial explosion in the training space of DON and make the efficient partial vertex order sampling , we employ a reinforcement learning model: the Policy Network, to adjust the partial order sampling probabilities during the training phase of DON automatically. To this end, the Policy Network can improve the training efficiency and guide DON to evolve towards a more effective model automatically. Comprehensive experiments on both synthetic and real data validate that DON-RL outperforms the current state-of-the-art heuristic algorithm consistently. Two case studies on graph compression and edge partitioning demonstrate the potential power of DON-RL in real applications.