The cold-start problem remains a significant challenge in recommendation systems, particularly for new users or unseen items with little to no historical data. Existing methods, including graph neural networks, often struggle in such scenarios. Inspired by the success of transformer models in natural language processing, we propose G-TRAC (Graph-Textual Representations Alignment for Cold-start Recommendations), a novel approach that integrates transformer-based textual modeling with graph neural networks. By effectively leveraging both textual and structural information, G-TRAC addresses cold-start challenges more effectively. Extensive experiments demonstrate its ability to enhance recommendation quality and generalize well across diverse scenarios.

Molecular representation learning plays a crucial role in advancing applications such as drug discovery and material design. Existing work leverages 2D and 3D modalities of molecular information for pre-training, aiming to capture comprehensive structural and geometric insights. However, these methods require paired 2D and 3D molecular data to train the model effectively and prevent it from collapsing into a single modality, posing limitations in scenarios where a certain modality is unavailable or computationally expensive to generate. To overcome this limitation, we propose FlexMol, a flexible molecule pre-training framework that learns unified molecular representations while supporting single-modality input. Specifically, inspired by the unified structure in vision-language models, our approach employs separate models for 2D and 3D molecular data, leverages parameter sharing to improve computational efficiency, and utilizes a decoder to generate features for the missing modality. This enables a multistage continuous learning process where both modalities contribute collaboratively during training, while ensuring robustness when only one modality is available during inference. Extensive experiments demonstrate that FlexMol achieves superior performance across a wide range of molecular property prediction tasks, and we also empirically demonstrate its effectiveness with incomplete data. Our code and data are available at https://github.com/tewiSong/FlexMol.

Graph self-supervised learning seeks to learn effective graph representations without relying on labeled data. Among various approaches, graph autoencoders (GAEs) have gained significant attention for their efficiency and scalability. Typically, GAEs take incomplete graphs as input and predict missing elements, such as masked node features or edges. Although effective, our experimental investigation reveals that traditional feature or edge masking paradigms primarily capture low-frequency signals in the graph and fail to learn expressive structural information. To address these issues, we propose Graph Positional Autoencoders (GraphPAE), which employ a dual-path architecture to reconstruct both node features and positions. Specifically, the feature path uses positional encoding to enhance the message-passing processing, improving the GAEs' ability to predict the corrupted information. The position path, on the other hand, leverages node representations to refine positions and approximate eigenvectors, thereby enabling the encoder to learn diverse frequency information. We conduct extensive experiments to verify the effectiveness of GraphPAE, including heterophilic node classification, graph property prediction, and transfer learning. The results demonstrate that GraphPAE achieves state-of-the-art performance and consistently outperforms the baselines by a large margin.

Text-attributed graphs (TAGs) have emerged as a powerful representation for modeling complex relationships across diverse domains. With the rise of large language models (LLMs), there is growing interest in leveraging their capabilities for graph learning. However, current approaches face significant challenges in embedding structural information into LLM-compatible formats, requiring either computationally expensive alignment mechanisms or manual graph verbalization techniques that often lose critical structural details. Moreover, these methods typically require labeled data from source domains for effective transfer learning, significantly constraining their adaptability. We propose STAG, a novel self-supervised framework that directly quantizes graph structural information into discrete tokens using a frozen codebook. Unlike traditional quantization approaches, our method employs soft assignment and KL divergence guided quantization to address the unique challenges of graph data, which lacks natural tokenization structures. Our framework enables both LLM-based and traditional learning approaches, supporting true zero-shot transfer learning without requiring labeled data even in the source domain. Extensive experiments demonstrate state-of-the-art performance across multiple node classification benchmarks while maintaining compatibility with different LLM architectures, offering an elegant solution to bridging graph learning with LLMs.

Understanding molecular structure and related knowledge is crucial for scientific research. Recent studies integrate molecular graphs with their textual descriptions to enhance molecular representation learning. However, they focus on the whole molecular graph and neglect frequently occurring subgraphs, known as motifs, which are essential for determining molecular properties. Without such fine-grained knowledge, these models struggle to generalize to unseen molecules and tasks that require motif-level insights. To bridge this gap, we propose FineMolTex, a novel Fine-grained Molecular graph-Text pre-training framework to jointly learn coarse-grained molecule-level knowledge and fine-grained motif-level knowledge. Specifically, FineMolTex consists of two pre-training tasks: a contrastive alignment task for coarse-grained matching and a masked multi-modal modeling task for fine-grained matching. In particular, the latter predicts the labels of masked motifs and words, which are selected based on their importance. By leveraging insights from both modalities, FineMolTex is able to understand the fine-grained matching between motifs and words. Finally, we conduct extensive experiments across three downstream tasks, achieving up to 230% improvement in the text-based molecule editing task. Additionally, our case studies reveal that FineMolTex successfully captures fine-grained knowledge, potentially offering valuable insights for drug discovery and catalyst design.

Chain-of-thought (CoT) prompting has achieved remarkable success in natural language processing (NLP). However, its vast potential remains largely unexplored for graphs. This raises an interesting question: How can we design CoT prompting for graphs to guide graph models to learn step by step? On one hand, unlike natural languages, graphs are non-linear and characterized by complex topological structures. On the other hand, many graphs lack textual data, making it difficult to formulate language-based CoT prompting. %Therefore we cannot directly adopt the CoT prompting methods used in the language domain. In this work, we propose the first CoT prompt learning framework for text-free graphs, GCoT. Specifically, we decompose the adaptation process for each downstream task into a series of inference steps, with each step consisting of prompt-based inference, ''thought'' generation, and thought-conditioned prompt learning. While the steps mimic CoT prompting in NLP, the exact mechanism differs significantly. Specifically, at each step, an input graph, along with a prompt, is first fed into a pre-trained graph encoder for prompt-based inference. We then aggregate the hidden layers of the encoder to construct a ''thought'', which captures the working state of each node in the current step. Conditioned on this thought, we learn a prompt specific to each node based on the current state. These prompts are fed into the next inference step, repeating the cycle. To evaluate and analyze the effectiveness of GCoT, we conduct comprehensive experiments on eight public datasets, which demonstrate the advantage of our approach.

Graphs are ubiquitous for modeling complex relationships between objects across various fields. Graph neural networks (GNNs) have become a mainstream technique for graph-based applications, but their performance heavily relies on abundant labeled data. To reduce labeling requirement, pre-training and prompt learning has become a popular alternative. However, most existing prompt methods do not distinguish between homophilic and heterophilic characteristics in graphs. In particular, many real-world graphs are non-homophilic-neither strictly nor uniformly homophilic-as they exhibit varying homophilic and heterophilic patterns across graphs and nodes. In this paper, we propose ProNoG, a novel pre-training and prompt learning framework for such non-homophilic graphs. First, we examineexisting graph pre-training methods, providing insights into the choice of pre-training tasks. Second, recognizing that each node exhibits unique non-homophilic characteristics, we propose a conditional network to characterize node-specific patterns in downstream tasks. Finally, we thoroughly evaluate and analyze ProNoG through extensive experiments on ten public datasets.

Modeling dynamic graphs, such as those found in social networks, recommendation systems, and e-commerce platforms, is crucial for capturing evolving relationships and delivering relevant insights over time. Traditional approaches primarily rely on graph neural networks with temporal components or sequence generation models, which often focus narrowly on the historical context of target nodes. This limitation restricts the ability to adapt to new and emerging patterns in dynamic graphs. To address this challenge, we propose a novel framework, Retrieval-Augmented Generation for Dy namic Graph modeling (RAG4DyG ), which enhances dynamic graph predictions by incorporating contextually and temporally relevant examples from broader graph structures. Our approach includes a time- and context-aware contrastive learning module to identify high-quality demonstrations and a graph fusion strategy to effectively integrate these examples with historical contexts. The proposed framework is designed to be effective in both transductive and inductive scenarios, ensuring adaptability to previously unseen nodes and evolving graph structures. Extensive experiments across multiple real-world datasets demonstrate the effectiveness of RAG4DyG in improving predictive accuracy and adaptability for dynamic graph modeling. The code and datasets are publicly available at https://github.com/YuxiaWu/RAG4DyG.

Learning effective latent representations for users and items is the cornerstone of recommender systems. Traditional approaches rely on user-item interaction data to map users and items into a shared latent space, but the limiting factor is often data sparsity. While reviews could mitigate this sparsity, existing review-aware recommendation models may not have fully exploited their potential. First, they typically rely heavily on reviews as side information or additional features, but reviews are not universal, with many users and items lacking them. Second, such approaches use reviews as supplementary information, allowing potential divergence or inconsistency between the review representations and the user-item space. To overcome these limitations, our work introduces a Review-centric Contrastive Alignment Framework for Recommendation (ReCAFR), which incorporates reviews into the core learning process, ensuring alignment among user, item, and review representations within a unified space. Specifically, we leverage two self-supervised contrastive strategies that not only exploit the review augmentation to alleviate sparsity, but also align the tripartite representations to enhance robustness. Empirical studies on public benchmark datasets demonstrate the effectiveness and robustness of ReCAFR

Online reviews allow consumers to provide detailed feedback on various aspects of items. Existing methods utilize these aspects to model users' fine-grained preferences for specific item features through graph neural networks. We argue that the performance of items on different aspects is important for making precise recommendations, which has not been taken into account by existing approaches, due to lack of data. In this paper, we propose an aspect performance-aware hypergraph neural network (APH) for the review-based recommendation, which learns the performance of items from the conflicting sentiment polarity of user reviews. Specifically, APH comprehensively models the relationships among users, items, aspects, and sentiment polarity by systematically constructing an aspect hypergraph based on user reviews. In addition, APH aggregates aspects representing users and items by employing an aspect performance-aware hypergraph aggregation method. It aggregates the sentiment polarities from multiple users by jointly considering user preferences and the semantics of their sentiments, determining the weights of sentiment polarities to infer the performance of items on various aspects. Such performances are then used as weights to aggregate neighboring aspects. Experiments on six real-world datasets demonstrate that APH improves MSE, Precision@5, and Recall@5 by an average of 2.30%, 4.89%, and 1.60% over the best baseline. The source code and data are available at https://github.com/dianziliu/APH.

Contemporary datasets sourced from the web often adopt a multiview format, collecting data from diverse sources, domains, or modules.Existing methodologies employed to analyze such datasets frequently overlook or inaccurately allocate the view-weights, pivotal metrics reflecting each view's significance.This work introduces EVA-MVC, a simple yet effective algorithm designed for Equitable View-weight Allocation (EVA) seamlessly integrated with arbitrary Multi-view Clustering (MVC) methods.Within the EVA module, we establish theoretical connections between view supplementarity and Multi-view Subspace Learning (MSL), leading to the partition of views into View Communities (VCs) based on these foundational principles.These VCs exhibit internal supplementarity similarities, facilitating Equitable View-weights Allocation through VCspecific MSL.The proposed EVA process precedes and operates independently of traditional or SOTA MVC approaches, requiring no additional processing or specialized design, making it an ideal preprocessing step for MVC applications.Through comprehensive evaluations across diverse multi-view datasets, our findings reveal that our EVA significantly enhances the effectiveness of mainstream MVC frameworks, resulting in a notable performance improvement.

Graphs are able to model interconnected entities in many online services, supporting a wide range of applications on the Web. This raises an important question: How can we train a graph foundational model on multiple source domains and adapt to an unseen target domain? A major obstacle is that graphs from different domains often exhibit divergent characteristics. Some studies leverage large language models to align multiple domains based on textual descriptions associated with the graphs, limiting their applicability to text-attributed graphs. For text-free graphs, a few recent works attempt to align different feature distributions across domains, while generally neglecting structural differences. In this work, we propose a novel Structure Alignment framework for text-free Multi-domain Graph Pre-Training and cross-domain adaptation (SAMGPT). It is designed to learn multi-domain knowledge from graphs originating in multiple source domains, which can then be adapted to address applications in an unseen target domain. Specifically, we introduce a set of structure tokens to harmonize structure-based aggregation across source domains during the pre-training phase. Next, for cross-domain adaptation, we design dual prompts, namely, holistic prompts and specific prompts, which adapt unified multi-domain structural knowledge and fine-grained, domain-specific information, respectively, to a target domain. Finally, we conduct comprehensive experiments on seven public datasets to evaluate and analyze the effectiveness of SAMGPT.

In the domain of Multi-view Subspace Clustering (MSC) in Latent Embedding Space (LES), existing methods aim to capture and leverage critical multi-view information by mapping it into a low-dimensional LES. However, several aspects can be further improved: (i) Fusion Strategy: Existing methods adopt either early fusion or late fusion to integrate multi-view information, limiting the effectiveness of the fusion. (ii) Diversity: Current methods often overlook the inherent diversity in the multi-view data by focusing on a single LES. (iii) Efficiency: LES-based methods exhibit high computational complexity, with cubic time and quadratic space requirements based on the number of samples. To address these issues, we propose a novel framework called MSC-DOLES (Multi-view Subspace Clustering in Diverse Orthogonal Latent Embedding Spaces), a novel framework designed to tackle these challenges. MSC-DOLES incorporates a two-stage fusion approach that generates and learns from multiple LES to maximize cross-view diversity. Orthogonality constraints on individual LES ensure view-internal diversity, resulting in a set of Diverse Orthogonal Latent Embedding Spaces (DOLES). The DOLES are then fused into a consensus anchor graph using learnable anchors. The final clustering is induced by partitioning the obtained graph without pre-processing. We develop an eight-step optimization algorithm for MSC-DOLES, which exhibits nearly linear time and space complexities relative to the number of samples. Extensive experiments demonstrate the superiority of MSC-DOLES over state-of-the-art methods.

Despite the success of conventional collaborative filtering (CF) approaches for recommendation systems, they exhibit limitations in leveraging semantic knowledge within the textual attributes of users and items. Recent focus on the application of large language models for recommendation (LLM4Rec) has highlighted their capability for effective semantic knowledge capture. However, these methods often overlook the collaborative signals in user behaviors. Some simply instruct-tune a language model, while others directly inject the embeddings of a CF-based model, lacking a synergistic fusion of different modalities. To address these issues, we propose a framework of Collaborative Cross-modal Fusion with Large Language Models, termed CCF-LLM, for recommendation. In this framework, we translate the user-item interactions into a hybrid prompt to encode both semantic knowledge and collaborative signals, and then employ an attentive cross-modal fusion strategy to effectively fuse latent embeddings of both modalities. Extensive experiments demonstrate that CCF-LLM outperforms existing methods by effectively utilizing semantic and collaborative signals in the LLM4Rec context.

Geodesic distances on manifolds have numerous applications in image processing, computer graphics and computer vision. In this work, we introduce an approach called 'LGGD' (Learned Generalized Geodesic Distances). This method involves generating node features by learning a generalized geodesic distance function through a training pipeline that incorporates training data, graph topology and the node content features. The strength of this method lies in the proven robustness of the generalized geodesic distances to noise and outliers. Our contributions encompass improved performance in node classification tasks, competitive results with state-of-the-art methods on real-world graph datasets, the demonstration of the learnability of parameters within the generalized geodesic equation on graph, and dynamic inclusion of new labels.

Link prediction is a fundamental task for graph analysis with important applications on the Web, such as social network analysis and recommendation systems, \etc\ Modern graph link prediction methods often employ a contrastive approach to learn robust node representations, where negative sampling is pivotal. Typical negative sampling methods aim to retrieve hard examples based on either predefined heuristics or automatic adversarial approaches, which might be inflexible or difficult to control. Furthermore, in the context of link prediction, most previous methods sample negative nodes from existing substructures of the graph, missing out on potentially more optimal samples in the latent space. To address these issues, we investigate a novel strategy of multi-level negative sampling that enables negative node generation with flexible and controllable "hardness'' levels from the latent space. Our method, called Conditional Diffusion-based Multi-level Negative Sampling (DMNS), leverages the Markov chain property of diffusion models to generate negative nodes in multiple levels of variable hardness and reconcile them for effective graph link prediction. We further demonstrate that DMNS follows the sub-linear positivity principle for robust negative sampling. Extensive experiments on several benchmark datasets demonstrate the effectiveness of DMNS.

Dynamic graph modeling is crucial for understanding complex structures in web graphs, spanning applications in social networks, recommender systems, and more. Most existing methods primarily emphasize structural dependencies and their temporal changes. However, these approaches often overlook detailed temporal aspects or struggle with long-term dependencies. Furthermore, many solutions overly complicate the process by emphasizing intricate module designs to capture dynamic evolutions. In this work, we harness the strength of the Transformer's self-attention mechanism, known for adeptly handling long-range dependencies in sequence modeling. Our approach offers a simple Transformer model, called SimpleDyG, tailored for dynamic graph modeling without complex modifications. We re-conceptualize dynamic graphs as a sequence modeling challenge and introduce a novel temporal alignment technique. This technique not only captures the inherent temporal evolution patterns within dynamic graphs but also streamlines the modeling process of their evolution. To evaluate the efficacy of SimpleDyG, we conduct extensive experiments on four real-world datasets from various domains. The results demonstrate the competitive performance of SimpleDyG in comparison to a series of state-of-the-art approaches despite its simple design.

Graph Neural Networks (GNNs) have emerged as a mainstream technique for graph representation learning. However, their efficacy within an end-to-end supervised framework is significantly tied to the availability of task-specific labels. To mitigate labeling costs and enhance robustness in few-shot settings, pre-training on self-supervised tasks has emerged as a promising method, while prompting has been proposed to further narrow the objective gap between pretext and downstream tasks. Although there has been some initial exploration of prompt-based learning on graphs, they primarily leverage a single pretext task, resulting in a limited subset of general knowledge that could be learned from the pre-training data. Hence, in this paper, we propose MultiGPrompt, a novel multi-task pre-training and prompting framework to exploit multiple pretext tasks for more comprehensive pre-trained knowledge. First, in pre-training, we design a set of pretext tokens to synergize multiple pretext tasks. Second, we propose a dual-prompt mechanism consisting of composed and open prompts to leverage task-specific and global pre-training knowledge, to guide downstream tasks in few-shot settings. Finally, we conduct extensive experiments on six public datasets to evaluate and analyze MultiGPrompt.

While the state-of-the-art network embedding approaches often learn high-quality embeddings for high-degree nodes with abundant structural connectivity, the quality of the embeddings for low-degree ortailnodes is often suboptimal due to their limited structural connectivity. While many real-world networks are long-tailed, to date little effort has been devoted to tail node embeddings. In this article, we formulate the goal of learning tail node embeddings as afew-shot regressionproblem, given the few links on each tail node. In particular, since each node resides in its own local context, we personalize the regression model for each tail node. To reduce overfitting in the personalization, we propose alocality-awaremeta-learning framework, calledmeta-tail2vec, which learns to learn the regression model for the tail nodes at different localities. Moreover, to address the heterogeneity in nodes and edges on heterogeneous information networks (HINs), we further extend the proposed model and formulatemeta-tail2vec+, which is based on a dual-adaptation mechanism to facilitate the locality-aware tail node embeddings on HINs. Finally, we conduct extensive experiments and demonstrate the promising results of both meta-tail2vec and its extension meta-tail2vec+.

Text classification is a fundamental problem in information retrieval with many real-world applications, such as predicting the topics of online articles and the categories of e-commerce product descriptions. However, low-resource text classification, with no or few labeled samples, presents a serious concern for supervised learning. Meanwhile, many text data are inherently grounded on a network structure, such as a hyperlink/citation network for online articles, and a user-item purchase network for e-commerce products. These graph structures capture rich semantic relationships, which can potentially augment low-resource text classification. In this paper, we propose a novel model called Graph-Grounded Pre-training and Prompting (G2P2) to address low-resource text classification in a two-pronged approach. During pre-training, we propose three graph interaction-based contrastive strategies to jointly pre-train a graph-text model; during downstream classification, we explore handcrafted discrete prompts and continuous prompt tuning for the jointly pre-trained model to achieve zero- and few-shot classification, respectively. Moreover, we explore the possibility of employing continuous prompt tuning for zero-shot inference. Specifically, we aim to generalize continuous prompts to unseen classes while leveraging a set of base classes. To this end, we extend G2P2 into G2P2$^*$, hinging on a new architecture of conditional prompt tuning. Extensive experiments on four real-world datasets demonstrate the strength of G2P2 in zero- and few-shot low-resource text classification tasks, and illustrate the advantage of G2P2$^*$in dealing with unseen classes.

Graphs can model complex relationships between objects, enabling a myriad of Web applications such as online page/article classification and social recommendation. While graph neural networks (GNNs) have emerged as a powerful tool for graph representation learning, in an end-to-end supervised setting, their performance heavily relies on a large amount of task-specific supervision. To reduce labeling requirement, the “pre-train, fine-tune” and “pre-train, prompt” paradigms have become increasingly common. In particular, prompting is a popular alternative to fine-tuning in natural language processing, which is designed to narrow the gap between pre-training and downstream objectives in a task-specific manner. However, existing study of prompting on graphs is still limited, lacking a universal treatment to appeal to different downstream tasks. In this paper, we proposeGraphPrompt, a novel pre-training and prompting framework on graphs.GraphPromptnot only unifies pre-training and downstream tasks into a common task template, but also employs a learnable prompt to assist a downstream task in locating the most relevant knowledge from the pre-trained model in a task-specific manner. In particular,GraphPromptadopts simple yet effective designs in both pre-training and prompt tuning: During pre-training, a link prediction-based task is used to materialize the task template; during prompt tuning, a learnable prompt vector is applied to theReadOutlayer of the graph encoder. To further enhanceGraphPromptin these two stages, we extend it intoGraphPrompt+with two major enhancements. First, we generalize a few popular graph pre-training tasks beyond simple link prediction to broaden the compatibility with our task template. Second, we propose a more generalized prompt design that incorporates a series of prompt vectors within every layer of the pre-trained graph encoder, in order to capitalize on the hierarchical information across different layers beyond just the readout layer. Finally, we conduct extensive experiments on five public datasets to evaluate and analyzeGraphPromptandGraphPrompt+.

Voucher abuse detection is an important anomaly detection problem in E-commerce. While many GNN-based solutions have emerged, the supervised paradigm depends on a large quantity of labeled data. A popular alternative is to adopt self-supervised pre-training using label-free data, and further fine-tune on a downstream task with limited labels. Nevertheless, the "pre-train, fine-tune" paradigm is often plagued by the objective gap between pre-training and downstream tasks. Hence, we propose VPGNN, a prompt-based fine-tuning framework on GNNs for voucher abuse detection. We design a novel graph prompting function to reformulate the downstream task into a similar template as the pretext task in pre-training, thereby narrowing the objective gap. Extensive experiments on both proprietary and public datasets demonstrate the strength of VPGNN in both few-shot and semi-supervised scenarios. Moreover, an online deployment of VPGNN in a production environment shows a 23.4% improvement over two existing deployed models.

Text classification is a fundamental problem in information retrieval with many real-world applications, such as predicting the topics of online articles and the categories of e-commerce product descriptions. However, low-resource text classification, with few or no labeled samples, poses a serious concern for supervised learning. Meanwhile, many text data are inherently grounded on a network structure, such as a hyperlink/citation network for online articles, and a user-item purchase network for e-commerce products. These graph structures capture rich semantic relationships, which can potentially augment low-resource text classification. In this paper, we propose a novel model called Graph-Grounded Pre-training and Prompting (G2P2) to address low-resource text classification in a two-pronged approach. During pre-training, we propose three graph interaction-based contrastive strategies to jointly pre-train a graph-text model; during downstream classification, we explore prompting for the jointly pre-trained model to achieve low-resource classification. Extensive experiments on four real-world datasets demonstrate the strength of G2P2 in zero- and few-shot low-resource text classification tasks

Graphs can model complex relationships between objects, enabling a myriad of Web applications such as online page/article classification and social recommendation. While graph neural networks (GNNs) have emerged as a powerful tool for graph representation learning, in an end-to-end supervised setting, their performance heavily relies on a large amount of task-specific supervision. To reduce labeling requirement, the “pre-train, fine-tune” and “pre-train, prompt” paradigms have become increasingly common. In particular, prompting is a popular alternative to fine-tuning in natural language processing, which is designed to narrow the gap between pre-training and downstream objectives in a task-specific manner. However, existing study of prompting on graphs is still limited, lacking a universal treatment to appeal to different downstream tasks. In this paper, we propose GraphPrompt, a novel pre-training and prompting framework on graphs. GraphPrompt not only unifies pre-training and downstream tasks into a common task template, but also employs a learnable prompt to assist a downstream task in locating the most relevant knowledge from the pre-trained model in a task-specific manner. Finally, we conduct extensive experiments on five public datasets to evaluate and analyze GraphPrompt.

On graph data, the multitude of node or edge types gives rise to heterogeneous information networks (HINs). To preserve the heterogeneous semantics on HINs, the rich node/edge types become a cornerstone of HIN representation learning. However, in real-world scenarios, type information is often noisy, missing or inaccessible. Assuming no type information is given, we define a so-called latent heterogeneous graph (LHG), which carries latent heterogeneous semantics as the node/edge types cannot be observed. In this paper, we study the challenging and unexplored problem of link prediction on an LHG. As existing approaches depend heavily on type-based information, they are suboptimal or even inapplicable on LHGs. To address the absence of type information, we propose a model named LHGNN, based on the novel idea of semantic embedding at node and path levels, to capture latent semantics on and between nodes. We further design a personalization function to modulate the heterogeneous contexts conditioned on their latent semantics w.r.t. the target node, to enable finer-grained aggregation. Finally, we conduct extensive experiments on four benchmark datasets, and demonstrate the superior performance of LHGNN.

Graph neural networks (GNNs) have witnessed widespread adoption due to their ability to learn superior representations for graph data. While GNNs exhibit strong discriminative power, they often fall short of learning the underlying node distribution for increased robustness. To deal with this, inspired by generative adversarial networks (GANs), we investigate the problem of adversarial learning on graph neural networks, and propose a novel framework named NAGNN (i.e., Neighbor-anchoring Adversarial Graph Neural Networks) for graph representation learning, which trains not only a discriminator but also a generator that compete with each other. In particular, we propose a novel neighbor-anchoring strategy, where the generator produces samples with explicit features and neighborhood structures anchored on a reference real node, so that the discriminator can perform neighborhood aggregation on the fake samples to learn superior representation. The advantage of our neighbor-anchoring strategy can be demonstrated both theoretically and empirically. Furthermore, as a by-product, our generator can synthesize realistic-looking features, enabling potential applications such as automatic content summarization. Finally, we conduct extensive experiments on four public benchmark datasets, and achieve promising results under both quantitative and qualitative evaluations.

While recommendation systems have been widely deployed, most existing approaches only capture user preferences in themacro-view, i.e., the user's general interest across all kinds of items. However, in real-world scenarios, user preferences could vary with items of different natures, which we call themicro-view. Both views are crucial for fully personalized recommendation, where an underpinning macro-view governs a multitude of finer-grained preferences in the micro-view. To model the dual views, in this paper, we propose a novel model called Dual-View Adaptive Recommendation (DVAR). In DVAR, we formulate the micro-view based on item categories, and further integrate it with the macro-view. Moreover, DVAR is designed to be adaptive, which is capable of automatically adapting to the dual-view preferences in response to different input users and item categories. To the best of our knowledge, this is the first attempt to integrate user preferences in macro- and micro- views in an adaptive way, without relying on additional side information such as text reviews. Finally, we conducted extensive quantitative and qualitative evaluations on several real-world datasets. Empirical results not only show that DVAR can significantly outperform other state-of-the-art recommendation systems, but also demonstrate the benefit and interpretability of the dual views.

SimRank is a popular link-based similarity measure on graphs. It enables a variety of applications with different modes of querying (e.g., single-pair, single-source and all-pair modes). In this paper, we propose UISim, a unified and incremental framework for all SimRank modes based on a scheduled approximation principle. UISim processes queries with incremental and prioritized exploration of the entire computation space, and thus allows flexible tradeoff of time and accuracy. On the other hand, it creates and shares common building blocks for online computation without relying on indexes, and thus is efficient to handle both static and dynamic graphs. Our experiments on various real-world graphs show that to achieve the same accuracy, UISim runs faster than its respective state-of-the-art baselines in each mode, and scales well on larger graphs.

Recommendation systems are popular in many domains. Researchers usually focus on the effectiveness of recommendation (e.g., precision) but neglect the popularity bias that may affect the fairness of the recommendation, which is also an important consideration that could influence the benefits of users and item providers. A few studies have been proposed to deal with the popularity bias, but they often face two limitations. Firstly, most studies only consider fairness for one side—either users or items, without achieving fairness jointly for both. Secondly, existing methods are not sufficiently tailored to each individual user or item to cope with the varying extent and nature of popularity bias. To alleviate these limitations, in this paper, we propose FAiR , a f airness-centric model that a dapt i vely mitigates the popularity bias in both users and items for r ecommendation. Concretely, we design explicit fairness discriminators to mitigate the popularity bias for each user and item locally, and an implicit discriminator to preserve fairness globally. Moreover, we dynamically adapt the model to different input users and items to handle the differences in their popularity bias. Finally, we conduct extensive experiments to demonstrate that our model significantly outperforms state-of-the-art baselines in fairness metrics, while remaining competitive in effectiveness.

While graph neural networks (GNNs) exhibit strong discriminative power, they often fall short of learning the underlying node distribution for increased robustness. To deal with this, inspired by generative adversarial networks (GANs), we investigate the problem of adversarial learning on graph neural networks, and propose a novel framework named NAGNN (i.e., Neighbor-anchoring Adversarial Graph Neural Networks) for graph representation learning, which trains not only a discriminator but also a generator that compete with each other. In particular, we propose a novel neighbor-anchoring strategy, where the generator produces samples with explicit features and neighborhood structures anchored on a reference real node, so that the discriminator can perform neighborhood aggregation on the fake samples to learn superior representations.

SimRank is a popular link-based similarity measure on graphs. It enables a variety of applications with different modes of querying. In this paper, we propose UISim, a unified and incremental framework for all SimRank modes based on a scheduled approximation principle. UISim processes queries with incremental and prioritized exploration of the entire computation space, and thus allows flexible tradeoff of time and accuracy. On the other hand, it creates and shares common “building blocks” for online computation without relying on indexes, and thus is efficient to handle both static and dynamic graphs. Our experiments on various real-world graphs show that to achieve the same accuracy, UISim runs faster than its respective state-of-the-art baselines, and scales well on larger graphs.

The prevalence of graph structures attracts a surge of investigation on graph data, enabling several downstream tasks such as multi-graph classification. However, in the multi-graph setting, graphs usually follow a long-tailed distribution in terms of their sizes, i.e., the number of nodes. In particular, a large fraction of tail graphs usually have small sizes. Though recent graph neural networks (GNNs) can learn powerful graph-level representations, they treat the graphs uniformly and marginalize the tail graphs which suffer from the lack of distinguishable structures, resulting in inferior performance on tail graphs. To alleviate this concern, in this paper we propose a novel graph neural network named SOLT-GNN, to close the representational gap between the head and tail graphs from the perspective of knowledge transfer. In particular, SOLT-GNN capitalizes on the co-occurrence substructures exploitation to extract the transferable patterns from head graphs. Furthermore, a novel relevance prediction function is proposed to memorize the pattern relevance derived from head graphs, in order to predict the complements for tail graphs to attain more comprehensive structures for enrichment. We conduct extensive experiments on five benchmark datasets, and demonstrate that our proposed model can outperform the state-of-the-art baselines.

Temporal graph representation learning has drawn significant attention for the prevalence of temporal graphs in the real world. However, most existing works resort to taking discrete snapshots of the temporal graph, or are not inductive to deal with new nodes, or do not model the exciting effects which is the ability of events to influence the occurrence of another event. In this work, We propose TREND, a novel framework for temporal graph representation learning, driven by TempoRal Event and Node Dynamics and built upon a Hawkes process-based graph neural network (GNN). TREND presents a few major advantages: (1) it is inductive due to its GNN architecture; (2) it captures the exciting effects between events by the adoption of the Hawkes process; (3) as our main novelty, it captures the individual and collective characteristics of events by integrating both event and node dynamics, driving a more precise modeling of the temporal process. Extensive experiments on four real-world datasets demonstrate the effectiveness of our proposed model.

Given that heterogeneous information networks (HIN) encompass nodes and edges belonging to different semantic types, they can model complex data in real-world scenarios. Thus, HIN embedding has received increasing attention, which aims to learn node representations in a low-dimensional space, in order to preserve the structural and semantic information on the HIN. In this regard, metagraphs, which model common and recurring patterns on HINs, emerge as a powerful tool to capture semantic-rich and often latent relationships on HINs. Although metagraphs have been employed to address several specific data mining tasks, they have not been thoroughly explored for the more general HIN embedding. In this paper, we leverage metagraphs to learn relationship-preserving HIN embedding in a self-supervised setting, to support various relationship mining tasks. In particular, we observe that most of the current approaches often under-utilize metagraphs, which are only applied in a pre-processing step and do not actively guide representation learning afterwards. Thus, we propose the novel framework of mg2vec, which learns the embeddings for metagraphs and nodes jointly. That is, metagraphs actively participates in the learning process by mapping themselves to the same embedding space as the nodes do. Moreover, metagraphs guide the learning through both first- and second-order constraints on node embeddings, to model not only latent relationships between a pair of nodes, but also individual preferences of each node. Finally, we conduct extensive experiments on three public datasets. Results show that mg2vec significantly outperforms a suite of state-of-the-art baselines in relationship mining tasks including relationship prediction, search and visualization.

While graph neural networks (GNNs) emerge as the state-of-the-art representation learning methods on graphs, they often require a large amount of labeled data to achieve satisfactory performance, which is often expensive or unavailable. To relieve the label scarcity issue, some pre-training strategies have been devised for GNNs, to learn transferable knowledge from the universal structural properties of the graph. However, existing pre-training strategies are only designed for homogeneous graphs, in which each node and edge belongs to the same type. In contrast, a heterogeneous graph embodies rich semantics, as multiple types of nodes interact with each other via different kinds of edges, which are neglected by existing strategies. In this paper, we propose a novel Contrastive Pre-Training strategy of GNNs on Heterogeneous Graphs (CPT-HG), to capture both the semantic and structural properties in a self-supervised manner. Specifically, we design semantic-aware pre-training tasks at both the relation- and subgraph-levels, and further enhance their representativeness by employing contrastive learning. We conduct extensive experiments on three real-world heterogeneous graphs, and promising results demonstrate the superior ability of our CPT-HG to transfer knowledge to various downstream tasks via pre-training.

Heterogeneous graphs (HGs), consisting of multiple types of nodes and links, can characterize a variety of real-world complex systems. Recently, heterogeneous graph neural networks (HGNNs), as a powerful graph embedding method to aggregate heterogeneous structure and attribute information, has earned a lot of attention. Despite the ability of HGNNs in capturing rich semantics which reveal different aspects of nodes, they still stay at a coarse-grained level which simply exploits structural characteristics. In fact, rich unstructured text content of nodes also carries latent but more fine-grained semantics arising from multi-facet topic-aware factors, which fundamentally manifest why nodes of different types would connect and form a specific heterogeneous structure. However, little effort has been devoted to factorizing them.

Graph neural networks (GNNs) emerge as the state-of-the-art representation learning methods on graphs and often rely on a large amount of labeled data to achieve satisfactory performance. Recently, in order to relieve the label scarcity issues, some works propose to pre-train GNNs in a self-supervised manner by distilling transferable knowledge from the unlabeled graph structures. Unfortunately, these pre-training frameworks mainly target at homogeneous graphs, while real interaction systems usually constitute large-scale heterogeneous graphs, containing different types of nodes and edges, which leads to new challenges on structure heterogeneity and scalability for graph pre-training. In this paper, we first study the problem of pre-training on large-scale heterogeneous graph and propose a novel pre-training GNN framework, named PT-HGNN. The proposed PT-HGNN designs both the node- and schema-level pre-training tasks to contrastively preserve heterogeneous semantic and structural properties as a form of transferable knowledge for various downstream tasks. In addition, a relationbased personalized PageRank is proposed to sparsify large-scale heterogeneous graph for efficient pre-training. Extensive experiments on one of the largest public heterogeneous graphs (OAG) demonstrate that our PT-HGNN significantly outperforms various state-of-the-art baselines.

The prevalence of graph structures in real-world scenarios enables important tasks such as node classification and link prediction. Graphs in many domains follow a long-tailed distribution in their node degrees, i.e., a significant fraction of nodes are tail nodes with a small degree. Although recent graph neural networks (GNNs) can learn powerful node representations, they treat all nodes uniformly and are not tailored to the large group of tail nodes. In particular, there is limited structural information (i.e., links) on tail nodes, resulting in inferior performance. Toward robust tail node embedding, in this paper we propose a novel graph neural network called Tail-GNN. It hinges on the novel concept of transferable neighborhood translation, to model the variable ties between a target node and its neighbors. On one hand, Tail-GNN learns a neighborhood translation from the structurally rich head nodes (i.e., high-degree nodes), which can be further transferred to the structurally limited tail nodes to enhance their representations. On the other hand, the ties with the neighbors are variable across different parts of the graph, and a global neighborhood translation is inflexible. Thus, we devise a node-wise adaptation to localize the global translation w.r.t. each node. Extensive experiments on five benchmark datasets demonstrate that our proposed Tail-GNN significantly outperforms the state-of-the-art baselines.

The 4th Workshop on Heterogeneous Information Network Analysis and Applications (HENA 2021) is co-located with the 27th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. The goal of this workshop is to bring together researchers and practitioners in the field and provide a forum for sharing new techniques and applications in heterogeneous information network analysis. This workshop has an exciting program that spans a number of subtopics, such as heterogeneous network embedding and graph neural networks, data mining techniques on heterogeneous information networks, and applications of heterogeneous information network analysis. The workshop program includes several invited speakers, lively discussion on emerging topics, and presentations of accepted research papers.

Semi-supervised node classification on graphs is an important research problem, with many real-world applications in information retrieval such as content classification on a social network and query intent classification on an e-commerce query graph. While traditional approaches are largely transductive, recent graph neural networks (GNNs) integrate node features with network structures, thus enabling inductive node classification models that can be applied to new nodes or even new graphs in the same feature space. However, inter-graph differences still exist across graphs within the same domain. Thus, training just one global model (e.g., a state-of-the-art GNN) to handle all new graphs, whilst ignoring the inter-graph differences, can lead to suboptimal performance. In this paper, we study the problem of inductive node classification across graphs. Unlike existing one-model-fits-all approaches, we propose a novel meta-inductive framework called MI-GNN to customize the inductive model to each graph under a meta-learning paradigm. That is, MI-GNN does not directly learn an inductive model; it learns the general knowledge of how to train a model for semi-supervised node classification on new graphs. To cope with the differences across graphs, MI-GNN employs a dual adaptation mechanism at both the graph and task levels. More specifically, we learn a graph prior to adapt for the graph-level differences, and a task prior to adapt for the task-level differences conditioned on a graph. Extensive experiments on five real-world graph collections demonstrate the effectiveness of our proposed model.

Real-world networks often exist with multiple views, where each view describes one type of interaction among a common set of nodes. For example, on a video-sharing network, while two user nodes are linked, if they have common favorite videos in one view, then they can also be linked in another view if they share common subscribers. Unlike traditional single-view networks, multiple views maintain different semantics to complement each other. In this article, we propose M ulti-view coll A borative N etwork E mbedding (MANE), a multi-view network embedding approach to learn low-dimensional representations. Similar to existing studies, MANE hinges on diversity and collaboration—while diversity enables views to maintain their individual semantics, collaboration enables views to work together. However, we also discover a novel form of second-order collaboration that has not been explored previously, and further unify it into our framework to attain superior node representations. Furthermore, as each view often has varying importance w.r.t. different nodes, we propose MANE , an attention -based extension of MANE, to model node-wise view importance. Finally, we conduct comprehensive experiments on three public, real-world multi-view networks, and the results demonstrate that our models consistently outperform state-of-the-art approaches.

In real-world problems, heterogeneous entities are often related to each other through multiple interactions, forming a Heterogeneous Interaction Graph (HIG). While modeling HIGs to deal with fundamental tasks, graph neural networks present an attractive opportunity that can make full use of the heterogeneity and rich semantic information by aggregating and propagating information from different types of neighborhoods. However, learning on such complex graphs, often with millions or billions of nodes, edges, and various attributes, could suffer from expensive time cost and high memory consumption. In this article, we attempt to accelerate representation learning on large-scale HIGs by adopting the importance sampling of heterogeneous neighborhoods in a batch-wise manner, which naturally fits with most batch-based optimizations. Distinct from traditional homogeneous strategies neglecting semantic types of nodes and edges, to handle the rich heterogeneous semantics within HIGs, we devise both type-dependent and type-fusion samplers where the former respectively samples neighborhoods of each type and the latter jointly samples from candidates of all types. Furthermore, to overcome the imbalance between the down-sampled and the original information, we respectively propose heterogeneous estimators including the self-normalized and the adaptive estimators to improve the robustness of our sampling strategies. Finally, we evaluate the performance of our models for node classification and link prediction on five real-world datasets, respectively. The empirical results demonstrate that our approach performs significantly better than other state-of-the-art alternatives, and is able to reduce the number of edges in computation by up to 93%, the memory cost by up to 92% and the time cost by up to 86%.

Data in the form of graphs are prevalent, ranging from biological and social networks to citation graphs and the Web. In particular, most real-world graphs are heterogeneous, containing objects of multiple types, which present new opportunities for many problems on graphs. Consider a typical proximity search problem on graphs, which boils down to measuring the proximity between two given nodes. Most earlier studies on homogeneous or bipartite graphs only measure a generic form of proximity, without accounting for different “semantic classes”-for instance, on a social network two users can be close for different reasons, such as being classmates or family members, which represent two distinct semantic classes. Learning these semantic classes are made possible on heterogeneous graphs through the concept of metagraphs. In this study, we identify metagraphs as a novel and effective means to characterize the common structures for a desired class of proximity. Subsequently, we propose a family of metagraph-based proximity, and employ a learning-to-rank technique that automatically learns the right parameters to suit the desired semantic class. In terms of efficiency, we develop a symmetry-based matching algorithm to speed up the computation of metagraph instances. Empirically, extensive experiments reveal that our metagraph-based proximity substantially outperforms the best competitor by more than 10 percent, and our matching algorithm can reduce matching time by more than half. As a further generalization, we aim to derive a general node and edge representation for heterogeneous graphs, in order to support arbitrary machine learning tasks beyond proximity search. In particular, we propose the finer-grained anchored metagraph, which is capable of discriminating the roles of nodes within the same metagraph. Finally, further experiments on the general representation show that we can outperform the state of the art significantly and consistently across various machine learning tasks.

Textual data is common and informative auxiliary information for recommender systems. Most prior art utilizes text for rating prediction, but rare work connects it to top-recommendation. Moreover, although advanced recommendation models capable of incorporating auxiliary information have been developed, none of these are specifically designed to model textual information, yielding a limited usage scenario for typical user-to-item recommendation. In this work, we present a framework of text-aware preference ranking (TPR) for top- recommendation, in which we comprehensively model the joint association of user-item interaction and relations between items and associated text. Using the TPR framework, we construct a joint likelihood function that explicitly describes two ranking structures: 1) item preference ranking (IPR) and 2) word relatedness ranking (WRR), where the former captures the item preference of each user and the latter captures the word relatedness of each item. As these two explicit structures are by nature mutually dependent, we propose TPR-OPT, a simple yet effective learning criterion that additionally includes implicit structures, such as relatedness between items and relatedness between words for each user for model optimization. Such a design not only successfully describes the joint association among users, words, and text comprehensively but also naturally yields powerful representations that are suitable for a range of recommendation tasks, including user-to-item, item-to-item, and user-to-word recommendation, as well as item-to-word reconstruction. In this paper, extensive experiments have been conducted on eight recommendation datasets, the results of which demonstrate that by including textual information from item descriptions, the proposed TPR model consistently outperforms state-of-the-art baselines on various recommendation tasks.

Network embedding is an active research area due to the prevalence of network-structured data. While the state of the art often learns high-quality embedding vectors for high-degree nodes with abundant structural connectivity, the quality of the embedding vectors for low-degree or tail nodes is often suboptimal due to their limited structural connectivity. While many real-world networks are long-tailed, to date little effort has been devoted to tail node embedding. In this paper, we formulate the goal of learning tail node embeddings as a few-shot regression problem, given the few links on each tail node. In particular, since each node resides in its own local context, we personalize the regression model for each tail node. To reduce overfitting in the personalization, we propose a locality-aware meta-learning framework, called meta-tail2vec, which learns to learn the regression model for the tail nodes at different localities. Finally, we conduct extensive experiments and demonstrate the promising results of meta-tail2vec. (Supplemental materials including code and data are available at https://github.com/smufang/meta-tail2vec.)

Cold-start recommendation has been a challenging problem due to sparse user-item interactions for new users or items. Existing efforts have alleviated the cold-start issue to some extent, most of which approach the problem at the data level. Earlier methods often incorporate auxiliary data as user or item features, while more recent methods leverage heterogeneous information networks (HIN) to capture richer semantics via higher-order graph structures. On the other hand, recent meta-learning paradigm sheds light on addressing cold-start recommendation at the model level, given its ability to rapidly adapt to new tasks with scarce labeled data, or in the context of cold-start recommendation, new users and items with very few interactions. Thus, we are inspired to develop a novel meta-learning approach named MetaHIN to address cold-start recommendation on HINs, to exploit the power of meta-learning at the model level and HINs at the data level simultaneously. The solution is non-trivial, for how to capture HIN-based semantics in the meta-learning setting, and how to learn the general knowledge that can be easily adapted to multifaceted semantics, remain open questions. In MetaHIN, we propose a novel semantic-enhanced tasks constructor and a co-adaptation meta-learner to address the two questions. Extensive experiments demonstrate that MetaHIN significantly outperforms the state of the arts in various cold-start scenarios. (Code and dataset are available at https://github.com/rootlu/MetaHIN.)

Recommender systems play an important role in helping users discover items of interest from a large resource collection in various online services. Although current deep neural network-based collaborative filtering methods have achieved state-of-the-art performance in recommender systems, they still face a few major weaknesses. Most importantly, such deep methods usually focus on the direct interaction between users and items only, without explicitly modeling high-order co-occurrence contexts. Furthermore, they treat the observed data uniformly, without fine-grained differentiation of importance or relevance in the user-item interactions and high-order co-occurrence contexts. Inspired by recent progress in memory networks, we propose a novel multiplex memory network for collaborative filtering (MMCF). More specifically, MMCF leverages a multiplex memory layer consisting of an interaction memory and two co-occurrence context memories simultaneously, in order to jointly capture and locate important and relevant information in both user-item interactions and co-occurrence contexts. Lastly, we conduct extensive experiments on four datasets, and the results show the superior performance of our model in comparison with a suite of state-of-the-art methods.

Network embedding effectively transforms complex network data into a low-dimensional vector space and has shown great performance in many real-world scenarios, such as link prediction, node classification, and similarity search. A plethora of methods have been proposed to learn node representations and achieve encouraging results. Nevertheless, little attention has been paid on the embedding technique for bipartite attributed networks, which is a typical data structure for modeling nodes from two distinct partitions.

Network embedding, which aims to represent network data in a low-dimensional space, has been commonly adopted for analyzing heterogeneous information networks (HIN). Although exiting HIN embedding methods have achieved performance improvement to some extent, they still face a few major weaknesses. Most importantly, they usually adopt negative sampling to randomly select nodes from the network, and they do not learn the underlying distribution for more robust embedding. Inspired by generative adversarial networks (GAN), we develop a novel framework HeGAN for HIN embedding, which trains both a discriminator and a generator in a minimax game. Compared to existing HIN embedding methods, our generator would learn the node distribution to generate better negative samples. Compared to GANs on homogeneous networks, our discriminator and generator are designed to be relation-aware in order to capture the rich semantics on HINs. Furthermore, towards more effective and efficient sampling, we propose a generalized generator, which samples "latent" nodes directly from a continuous distribution, not confined to the nodes in the original network as existing methods are. Finally, we conduct extensive experiments on four real-world datasets. Results show that we consistently and significantly outperform state-of-the-art baselines across all datasets and tasks.

We study the important problem of user alignment in e-commerce: to predict whether two online user identities that access an e-commerce site from different devices belong to one real-world person. As input, we have a set of user activity logs from Taobao and some labeled user identity linkages. User activity logs can be modeled using a heterogeneous interaction graph (HIG), and subsequently the user alignment task can be formulated as a semi-supervised HIG embedding problem. HIG embedding is challenging for two reasons: its heterogeneous nature and the presence of edge features. To address the challenges, we propose a novel Heterogeneous Embedding Propagation (HEP) model. The core idea is to iteratively reconstruct a node's embedding from its heterogeneous neighbors in a weighted manner, and meanwhile propagate its embedding updates from reconstruction loss and/or classification loss to its neighbors. We conduct extensive experiments on large-scale datasets from Taobao, demonstrating that HEP significantly outperforms state-of-the-art baselines often by more than 10% in F-scores.

Given ubiquitous graph data such as the Web and social networks, proximity search on graphs has been an active research topic. The task boils down to measuring the proximity between two nodes on a graph. Although most earlier studies deal with homogeneous or bipartite graphs only, many real-world graphs are heterogeneous with objects of various types, giving rise to different semantic classes of proximity. For instance, on a social network two users can be close for different reasons, such as being classmates or family members, which represent two distinct classes of proximity. Thus, it becomes inadequate to only measure a “generic” form of proximity as previous works have focused on. In this paper, we identify metagraphs as a novel and effective means to characterize the common structures for a desired class of proximity. Subsequently, we propose a family of metagraph-based proximity, and employ a supervised technique to automatically learn the right form of proximity within its family to suit the desired class. As it is expensive to match (i.e., find the instances of) a metagraph, we propose the novel approaches of dual-stage training and symmetry-based matching to speed up. Finally, our experiments reveal that our approach is significantly more accurate and efficient. For accuracy, we outperform the baselines by 11% and 16% in NDCG and MAP, respectively. For efficiency, dual-stage training reduces the overall matching cost by 83%, and symmetry-based matching further decreases the cost of individual metagraphs by 52%.

As the Web hosts rich information about real-world entities, our information quests become increasingly entity centric. In this paper, we study the problem of focused harvesting of Web pages for entity aspects, to support downstream applications such as business analytics and building a vertical portal. Given that search engines are the de facto gateways to assess information on the Web, we recognize the essence of our problem as Learning to Query (L2Q) - to intelligently select queries so that we can harvest pages, via a search engine, focused on an entity aspect of interest. Thus, it is crucial to quantify the utilities of the candidate queries w.r.t. some entity aspect. In order to better estimate the utilities, we identify two opportunities and address their challenges. First, a target entity in a given domain has many peers. We leverage these peer entities to become domain aware. Second, a candidate query may “overlap” with the past queries that have already been fired. We account for these past queries to become context aware. Empirical results show that our approach significantly outperforms both algorithmic and manual baselines by 16% and 10% in F-scores, respectively.

Graph-based proximity has many applications with different ranking needs. However, most previous works only stress the sense of importance by finding “popular” results for a query. Often times important results are overly general without being well-tailored to the query, lacking a sense of specificity - which only emerges recently. Even then, the two senses are treated independently, and only combined empirically. In this paper, we generalize the well-studied importance-based random walk into a round trip and develop RoundTripRank, seamlessly integrating specificity and importance in one coherent process. We also recognize the need for a flexible trade-off between the two senses, and further develop RoundTripRank+ based on a scheme of hybrid random surfers. For efficient computation, we start with a basic model that decomposes RoundTripRank into smaller units. For each unit, we apply a novel two-stage bounds updating framework, enabling an online top-K algorithm 2SBound. Finally, our experiments show that RoundTripRank and RoundTripRank+ are robust over various ranking tasks, and 2SBound enables scalable online processing.

As Personalized PageRank has been widely leveraged for ranking on a graph, the efficient computation of Personalized PageRank Vector (PPV) becomes a prominent issue. In this paper, we propose FastPPV, an approximate PPV computation algorithm that is incremental and accuracy-aware. Our approach hinges on a novel paradigm of scheduled approximation: the computation is partitioned and scheduled for processing in an "organized" way, such that we can gradually improve our PPV estimation in an incremental manner, and quantify the accuracy of our approximation at query time. Guided by this principle, we develop an efficient hub based realization, where we adopt the metric of hub-length to partition and schedule random walk tours so that the approximation error reduces exponentially over iterations. Furthermore, as tours are segmented by hubs, the shared substructures between different tours (around the same hub) can be reused to speed up query processing both within and across iterations. Finally, we evaluate FastPPV over two real-world graphs, and show that it not only significantly outperforms two state-of-the-art baselines in both online and offline phrases, but also scale well on larger graphs. In particular, we are able to achieve near-constant time online query processing irrespective of graph size.

Conventional classification methods tend to focus on features of individual objects, while missing out on potentially valuable pairwise features that capture the relationships between objects. Although recent developments on graph regularization exploit this aspect, existing works generally assume only a single kind of pairwise feature, which is often insufficient. We observe that multiple, heterogeneous pairwise features can often complement each other and are generally more robust in modeling the relationships between objects. Furthermore, as some objects are easier to classify than others, objects with higher initial classification confidence should be weighed more towards classifying related but more ambiguous objects, an observation missing from previous graph regularization techniques. In this paper, we propose a Dirichlet-based regularization framework that supports the combination of heterogeneous pairwise features with confidence-aware prediction using limited labeled training data. Next, we showcase a few applications of our framework in information retrieval, focusing on the problem of query intent classification. Finally, we demonstrate through a series of experiments the advantages of our framework on a large-scale real-world dataset.

While tuple extraction for a given relation has been an active research area, its dual problem of pattern search-- to find and rank patterns in a principled way-- has not been studied explicitly. In this paper, we propose and address the problem of pattern search, in addition to tuple extraction. As our objectives, we stress reusability for pattern search and scalability of tuple extraction, such that our approach can be applied to very large corpora like the Web. As the key foundation, we propose a conceptual model PRDualRank to capture the notion of precision and recall for both tuples and patterns in a principled way, leading to the "rediscovery" of the Pattern-Relation Duality-- the formal quantification of the reinforcement between patterns and tuples with the metrics of precision and recall. We also develop a concrete framework for PRDualRank, guided by the principles of a perfect sampling process over a complete corpus. Finally, we evaluated our framework over the real Web. Experiments show that on all three target relations our principled approach greatly outperforms the previous state-of-the-art system in both effectiveness and efficiency. In particular, we improved optimal F-score by up to 64%.