Explainable AI (XAI) refers to techniques that provide human-understandable insights into the workings of AI models. Recently, the focus of XAI has been extended toward explaining Large Language Models (LLMs). This extension calls for a significant transformation in the XAI methodologies for two reasons. First, many existing XAI methods cannot be directly applied to LLMs due to their complexity and advanced capabilities. Second, as LLMs are increasingly deployed in diverse applications, the role of XAI shifts from merely opening the “black box” to actively enhancing the productivity and applicability of LLMs in real-world settings. Meanwhile, the conversation and generation abilities of LLMs can reciprocally enhance XAI. Therefore, in this article, we introduce Usable XAI in the context of LLMs by analyzing (1) how XAI can explain and improve LLM-based AI systems and (2) how XAI techniques can be improved by using LLMs. We introduce 10 strategies, introducing the key techniques for each and discussing their associated challenges. We also provide case studies to demonstrate how to obtain and leverage explanations.

Modern text classification methods heavily rely on contextual embeddings from large language models (LLMs). Compared to human-engineered features, these embeddings provide automatic and effective representations for classification model training. However, they also introduce a challenge: we lose the ability to manually remove unintended features, such as sensitive or task-irrelevant features, to guarantee regulatory compliance or improve the generalizability of classification models. This limitation arises because LLM embeddings are opaque and difficult to interpret. In this paper, we propose a novel framework to identify and regularize unintended features in the LLM latent space. Specifically, we first pre-train a sparse autoencoder (SAE) to extract interpretable features from LLM latent spaces. To ensure the SAE can capture task-specific features, we further fine-tune it on task-specific datasets. In training the classification model, we propose a simple and effective regularizer, by minimizing the similarity between the classifier weights and the identified unintended feature, to remove the impact of these unintended features on classification. We evaluate the proposed framework on three real-world tasks, including toxic chat detection, reward modeling, and disease diagnosis. Results show that the proposed self-regularization framework can improve the classifier's generalizability by regularizing those features that are not semantically correlated to the task. This work pioneers controllable text classification on LLM latent spaces by leveraging interpreted features to address generalizability, fairness, and privacy challenges. The code and data are publicly available at https://github.com/JacksonWuxs/Controllable_LLM_Classifier.

Representation learning on text-attributed graphs (TAGs), where nodes are associated with textual descriptions, is crucial for textual and relational knowledge systems, such as social media and recommendation scenarios. However, state-of-the-art embedding methods for TAGs primarily focus on fine-tuning pre-trained language models (PLMs) using structure-aware training objectives. While effective, these methods are tailored for individual TAG and cannot generalize across various graph scenarios. Given the shared textual space, leveraging multiple TAGs for joint fine-tuning, aligning text and graph structure from different aspects, would be more beneficial. Therefore, we propose the Unified Graph Language Model (UniGLM), a novel foundation model pretrained over multiple TAGs from a variety of domains, which can generalize well to both in-domain and cross-domain graph scenarios. Specifically, UniGLM fine-tunes well-established PLMs (e.g., Sentence-BERT) using a domain-aware contrastive learning objective that unifies structure heterogeneity and node statistics across various domains with an adaptive and learnable positive sample selection scheme. Additionally, a lazy updating module is introduced to speed up training by reducing repetitive encoding of positive samples. Extensive datasets across multiple domains, downstream tasks (node classification and link prediction), and a spectrum of graph backbones (supervised and self-supervised graph models) are conducted to compare UniGLM with state-of-the-art baselines. Our empirical observations suggest that UniGLM can generate informative representations for cross-domain graphs observed in the training. More importantly, UniGLM also exhibits competitive transfer ability in encoding unseen TAGs that are not used for training. This study provides deep insights into how to adapt PLMs to graph data and demonstrates the potential of building foundation model for graph representation learning.

As Graph Neural Networks (GNNs) have been widely used in real-world applications, model explanations are required not only by users but also by legal regulations. However, simultaneously achieving high fidelity and low computational costs in generating explanations has been a challenge for current methods. In this work, we propose a framework of GNN explanation named L e A rn R emoval-based A ttribution (LARA) to address this problem. Specifically, we introduce removal-based attribution and demonstrate its substantiated link to interpretability fidelity theoretically and experimentally. The explainer in LARA learns to generate removal-based attribution which enables providing explanations with high fidelity. A strategy of subgraph sampling is designed in LARA to improve the scalability of the training process. In the deployment, LARA can efficiently generate the explanation through a feed-forward pass. We benchmark our approach with other state-of-the-art GNN explanation methods on six datasets. Results highlight the effectiveness of our framework regarding both efficiency and fidelity. In particular, LARA is 3.1 \(\times\) faster and achieves higher fidelity than the state-of-the-art method on the large dataset ogbn-arxiv (more than 160K nodes and 1M edges), showing its great potential in real-world applications. Our source code is available at https://github.com/yaorong0921/LARA .

Currently, attention mechanism has become a standard fixture in most state-of-the-art natural language processing (NLP) models, not only due to the outstanding performance it could gain but also due to plausible innate explanations for the behaviors of neural architectures it provides, which is notoriously difficult to analyze. However, recent studies show that attention is unstable against randomness and perturbations during training or testing, such as random seeds and slight perturbation of embedding vectors, which impedes it from becoming a faithful explanation tool. Thus, a natural question is whether we can find some substitute for the current attention that is more stable and could keep the most important characteristics of explanation and prediction of attention. In this paper, to resolve the problem, we provide a rigorous definition of such alternate namely SEAT (Stable andExplainableAttention). Specifically, a SEAT should have the following three properties: (1) Its prediction distribution is enforced to be close to the distribution based on the vanilla attention; (2) Its top-$k$indices have large overlaps with those of the vanilla attention; (3) It is robust w.r.t perturbations, i.e., any slight perturbation on SEAT will not change the prediction distribution too much, which implicitly indicates that it is stable to randomness and perturbations. To further improve the interpretability stability against perturbations, based on SEAT we provide another definition called SEAT++. Then we propose a method to get a SEAT++, which could be considered an ad hoc modification for canonical attention. Finally, through intensive experiments on various datasets, we compare our SEAT and SEAT++ with other baseline methods using RNN, BiLSTM, and BERT architectures via six different evaluation metrics for model interpretation, stability, and accuracy. Results show that SEAT and SEAT++ are more stable against different perturbations and randomness while also keeping the explainability of attention, which indicates they provide more faithful explanations. Moreover, compared with vanilla attention, there is almost no utility (accuracy) degradation for SEAT and SEAT++.

Large Language Models (LLMs) have shown proficiency in question-answering tasks but often struggle to integrate real-time knowledge, leading to potentially outdated or inaccurate responses. This problem becomes even more challenging when dealing with multi-hop questions, since they require LLMs to update and integrate multiple knowledge pieces relevant to the questions. To tackle the problem, we propose the Retrieval-Augmented model Editing (RAE) framework for multi-hop question answering. RAE first retrieves edited facts and then refines the language model through in-context learning. Specifically, our retrieval approach, based on mutual information maximization, leverages the reasoning abilities of LLMs to identify chain facts that traditional similarity-based searches might miss. In addition, our framework includes a pruning strategy to eliminate redundant information from the retrieved facts, which enhances the editing accuracy and mitigates the hallucination problem. Our framework is supported by theoretical justification for its fact retrieval efficacy. Finally, comprehensive evaluation across various LLMs validates RAE's ability in providing accurate answers with updated knowledge. Our code is available at: https://github.com/sycny/RAE.

Recommending products to users with intuitive explanations helps improve the system in transparency, persuasiveness, and satisfaction. Existing interpretation techniques include post hoc methods and interpretable modeling. The former category could quantitatively analyze input contribution to model prediction but has limited interpretation faithfulness, while the latter could explain model internal mechanisms but may not directly attribute model predictions to input features. In this study, we propose a novel Dual Interpretable Recommendation model called DIRECT, which integrates ideas of the two interpretation categories to inherit their advantages and avoid limitations. Specifically, DIRECT makes use of item descriptions as explainable evidence for recommendation. First, similar to the post hoc interpretation, DIRECT could attribute the prediction of a user preference score to textual words of the item descriptions. The attribution of each word is related to its sentiment polarity and word importance, where a word is important if it corresponds to an item aspect that the user is interested in. Second, to improve the interpretability of embedding space, we propose to extract high-level concepts from embeddings, where each concept corresponds to an item aspect. To learn discriminative concepts, we employ a concept bottleneck layer and maximize the coding rate reduction on word-aspect embeddings by leveraging a word–word affinity graph extracted from a pre-trained language model. In this way, DIRECT simultaneously achieves faithful attribution and usable interpretation of embedding space. We also show that DIRECT achieves linear inference time complexity regarding the length of item reviews. We conduct experiments including ablation studies on five real-world datasets. Quantitative analysis, visualizations, and case studies verify the interpretability of DIRECT. Our code is available at: https://github.com/JacksonWuxs/DIRECT .

Large language models (LLMs) have demonstrated impressive capabilities in natural language processing. However, their internal mechanisms are still unclear and this lack of transparency poses unwanted risks for downstream applications. Therefore, understanding and explaining these models is crucial for elucidating their behaviors, limitations, and social impacts. In this article, we introduce a taxonomy of explainability techniques and provide a structured overview of methods for explaining Transformer-based language models. We categorize techniques based on the training paradigms of LLMs: traditional fine-tuning-based paradigm and prompting-based paradigm. For each paradigm, we summarize the goals and dominant approaches for generating local explanations of individual predictions and global explanations of overall model knowledge. We also discuss metrics for evaluating generated explanations and discuss how explanations can be leveraged to debug models and improve performance. Lastly, we examine key challenges and emerging opportunities for explanation techniques in the era of LLMs in comparison to conventional deep learning models.

The application of recommendation technologies has been crucial in the promotion of physical and digital content across numerous global platforms such as Amazon, Apple, and Netflix. Our study aims to investigate the advantages of employing recommendation technologies on educational platforms, with a particular focus on an educational platform for learning and practicing music. Our research is based on data from Tomplay, a music platform that offers sheet music with professional audio recordings, enabling users to discover and practice music content at varying levels of difficulty. Through our analysis, we emphasize the distinct interaction patterns on educational platforms like Tomplay, which we compare with other commonly used recommendation datasets. We find that interactions are comparatively sparse on educational platforms, with users often focusing on specific content as they learn, rather than interacting with a broader range of material. Therefore, our primary goal is to address the issue of data sparsity. We achieve this through entity resolution principles and propose a neural network (NN)-based recommendation model. Further, we improve this model by utilizing graph neural networks (GNNs), which provide superior predictive accuracy compared to NNs. Notably, our study demonstrates that GNNs are highly effective even for users with little or no historical preferences (cold-start problem). Our cold-start experiments also provide valuable insights into an independent issue, namely, the number of historical interactions needed by a recommendation model to gain a comprehensive understanding of a user. Our findings demonstrate that a platform acquires a solid knowledge of a user’s general preferences and characteristics with 50 past interactions. Overall, our study makes significant contributions to information systems research on business analytics and prescriptive analytics. Moreover, our framework and evaluation results offer implications for various stakeholders, including online educational institutions, education policymakers, and learning platform users.

Backdoor attacks inject poisoned samples into training data, where backdoor triggers are embedded into the model trained on the mixture of poisoned and clean samples.An interesting phenomenon can be observed in the training process: the loss of poisoned samples tends to drop significantly faster than that of clean samples, which we call the early-fitting phenomenon.Early-fitting provides a simple but effective evidence to defend against backdoor attacks, where the poisoned samples can be detected by selecting the samples with the lowest loss values in the early training epochs.Then, two questions naturally arise: (1) What characteristics of poisoned samples cause early-fitting?(2) Does a stronger attack exist which could circumvent the defense methods?To answer the first question, we find that early-fitting could be attributed to a unique property among poisoned samples called synchronization, which depicts the similarity between two samples at different layers of a model.Meanwhile, the degree of synchronization could be controlled based on whether it is captured by shallow or deep layers of the model.Then, we give an affirmative answer to the second question by proposing a new backdoor attack method, Deep Backdoor Attack (DBA), which utilizes deep synchronization to reverse engineer trigger patterns by activating neurons in the deep layer of a base neural network.Experimental results validate our propositions and the effectiveness of DBA.Our code is available at https://github.com/GuanZihan/Deep-Backdoor-Attack.

Self-supervised learning with masked autoencoders has recently gained popularity for its ability to produce effective image or textual representations, which can be applied to various downstream tasks without retraining. However, we observe that the current masked autoencoder models lack good generalization ability on graph data. To tackle this issue, we propose a novel graph masked autoencoder framework called GiGaMAE. Different from existing masked autoencoders that learn node presentations by explicitly reconstructing the original graph components (e.g., features or edges), in this paper, we propose to collaboratively reconstruct informative and integrated latent embeddings. By considering embeddings encompassing graph topology and attribute information as reconstruction targets, our model could capture more generalized and comprehensive knowledge. Furthermore, we introduce a mutual information based reconstruction loss that enables the effective reconstruction of multiple targets. This learning objective allows us to differentiate between the exclusive knowledge learned from a single target and common knowledge shared by multiple targets. We evaluate our method on three downstream tasks with seven datasets as benchmarks. Extensive experiments demonstrate the superiority of GiGaMAE against state-of-the-art baselines. We hope our results will shed light on the design of foundation models on graph-structured data. Our code is available at: https://github.com/sycny/GiGaMAE.

The recent breakthrough achieved by graph neural networks (GNNs) with few labeled data accelerates the pace of deploying GNNs on real-world applications. While several efforts have been made to scale GNNs training for large-scale graphs, GNNs still suffer from the scalability challenge of model inference, due to the graph dependency issue incurred by the message-passing mechanism, therefore hindering its deployment in resource-constrained applications. An intuitive remedy is compressing the cumbersome GNN model into inference-friendly multi-layer perceptrons (MLPs) using knowledge distillation (KD). However, the standard KD strategy, i.e., training MLPs using the soft labels of labeled and unlabeled nodes from the teacher, is suboptimal, since the GNN teacher would inevitably make wrong predictions for unlabeled data, especially in the semi-supervised scenario. To address this, we propose a novel Reliable Knowledge Distillation framework for MLP optimization (RKDMLP), which shows strong promise in achieving a “sweet point” in co-optimizing model accuracy and efficiency. Its core insight is to use a meta-policy to filter out those unreliable soft labels. To train the meta-policy, we design a reward-driven objective based on a meta-set and adopt policy gradient to optimize the expected reward. Then we apply the meta-policy to the unlabeled nodes and select the most reliable soft labels for distillation. Extensive experiments across various GNN backbones, on 7 small graphs and 2 large-scale datasets from the challenging Open Graph Benchmark, demonstrate the superiority of our proposal. Moreover, RKD-MLP also shows good robustness iv.r.t. graph topology and node feature noises.

Graph neural networks (GNNs), which learn the node representations by recursively aggregating information from its neighbors, have become a predominant computational tool in many domains. To handle large-scale graphs, most of the existing methods partition the input graph into multiple sub-graphs (e.g., through node clustering) and apply batch training to save memory cost. However, such batch training will lead to label bias within each batch and result in over- confidence in model predictions. Since the connected nodes with positively related labels tend to be assigned together, the traditional cross-entropy minimization process attends on the predictions of biased classes at a batch to intensify the overfitting issue. To overcome the problem of label bias, we propose adaptive label smoothing (ALS) method to replace the one-hot hard labels with smoothed ones, which learns to allocate label confidences from the biased classes to the others. Specifically, ALS propagates node labels to aggregate the neighborhood label distribution in a pre-processing step, and then updates the optimal smoothed labels online to adapt to specific graph structure. Experiments on the real-world datasets demonstrate that ALS can be generally applied to the main scalable learning frameworks to calibrate the biased labels and improve generalization performances.

A knowledge graph (KG) consists of numerous triples, in which each triple, i.e., (head entity, relation, tail entity), denotes a real-world assertion. Many large-scale KGs have been developed, e.g., general-purpose KGs Freebase and YAGO. Also, lots of domain-specific KGs are emerging, e.g., COVID-19 KGs, biomedical KGs, and agricultural KGs. By embedding KGs into low-dimensional vectors, i.e., representations of entities and relations, we could integrate KGs into machine learning models and enhance the performance of many prediction tasks, including search, recommendations, and question answering. During the construction, refinement, embedding, and application of KGs, a variety of KG learning algorithms have been developed to handle challenges in various real-world scenarios. Moreover, graph neural networks have also brought new opportunities to KG learning. This workshop aims to engage with active researchers from KG communities, recommendation communities, natural language processing communities, and other communities, and deliver state-of-the-art research insights into the core challenges in KG learning.

Self-supervised learning (SSL) has been demonstrated to be effective in pre-training models that can be generalized to various downstream tasks. Graph Autoencoder (GAE), an increasingly popular SSL approach on graphs, has been widely explored to learn node representations without ground-truth labels. However, recent studies show that existing GAE methods could only perform well on link prediction tasks, while their performance on classification tasks is rather limited. This limitation casts doubt on the generalizability and adoption of GAE. In this paper, for the first time, we show that GAE can generalize well to both link prediction and classification scenarios, including node-level and graph-level tasks, by redesigning its critical building blocks from the graph masking perspective. Our proposal is called Self-Supervised Graph Autoencoder--S2GAE, which unleashes the power of GAEs with minimal yet nontrivial efforts. Specifically, instead of reconstructing the whole input structure, we randomly mask a portion of edges and learn to reconstruct these missing edges with an effective masking strategy and an expressive decoder network. Moreover, we theoretically prove that S2GAE could be regarded as an edge-level contrastive learning framework, providing insights into why it generalizes well. Empirically, we conduct extensive experiments on 21 benchmark datasets across link prediction and node & graph classification tasks. The results validate the superiority of S2GAE against state-of-the-art generative and contrastive methods. This study demonstrates the potential of GAE as a universal representation learner on graphs. Our code is publicly available at https://github.com/qiaoyu-tan/S2GAE.

Graph neural networks (GNNs) have received remarkable success in link prediction (GNNLP) tasks. Existing efforts first predefine the subgraph for the whole dataset and then apply GNNs to encode edge representations by leveraging the neighborhood structure induced by the fixed subgraph. The prominence of GNNLP methods significantly relies on the adhoc subgraph. Since node connectivity in real-world graphs is complex, one shared subgraph is limited for all edges. Thus, the choices of subgraphs should be personalized to different edges. However, performing personalized subgraph selection is nontrivial since the potential selection space grows exponentially to the scale of edges. Besides, the inference edges are not available during training in link prediction scenarios, so the selection process needs to be inductive. To bridge the gap, we introduce a Personalized Subgraph Selector (PS2) as a plug-and-play framework to automatically, personally, and inductively identify optimal subgraphs for different edges when performing GNNLP. PS2 is instantiated as a bi-level optimization problem that can be efficiently solved differently. Coupling GNNLP models with PS2, we suggest a brand-new angle towards GNNLP training: by first identifying the optimal subgraphs for edges; and then focusing on training the inference model by using the sampled subgraphs. Comprehensive experiments endorse the effectiveness of our proposed method across various GNNLP backbones (GCN, GraphSage, NGCF, LightGCN, and SEAL) and diverse benchmarks (Planetoid, OGB, and Recommendation datasets). Our code is publicly available at https://github.com/qiaoyu-tan/PS2

Machine learning models are becoming pervasive in high-stakes applications. Despite their clear benefits in terms of performance, the models could show discrimination against minority groups and result in fairness issues in a decision-making process, leading to severe negative impacts on the individuals and the society. In recent years, various techniques have been developed to mitigate the unfairness for machine learning models. Among them, in-processing methods have drawn increasing attention from the community, where fairness is directly taken into consideration during model design to induce intrinsically fair models and fundamentally mitigate fairness issues in outputs and representations. In this survey, we review the current progress of in-processing fairness mitigation techniques. Based on where the fairness is achieved in the model, we categorize them into explicit and implicit methods, where the former directly incorporates fairness metrics in training objectives, and the latter focuses on refining latent representation learning. Finally, we conclude the survey with a discussion of the research challenges in this community to motivate future exploration.

Graph anomaly detection (GAD) has wide applications in real-world networked systems. In many scenarios, people need to identify anomalies on new (sub)graphs, but they may lack labels to train an effective detection model. Since recent semi-supervised GAD methods, which can leverage the available labels as prior knowledge, have achieved superior performance than unsupervised methods, one natural idea is to directly adopt a trained semi-supervised GAD model to the new (sub)graphs for testing. However, we find that existing semi-supervised GAD methods suffer from poor generalization issues, i.e., well-trained models could not perform well on an unseen area (i.e., not accessible in training) of the graph. Motivated by this, we formally define the problem of generalized graph anomaly detection that aims to effectively identify anomalies on both the training-domain graph(s) and the unseen test graph(s). Nevertheless, it is a challenging task since only limited labels are available, and the normal data distribution may differ between training and testing data. Accordingly, we propose a data augmentation method namedAugAN(Augmentation forAnomaly andNormal distributions) to enrich training data and adopt a customized episodic training strategy for learning with the augmented data. Extensive experiments verify the effectiveness ofAugANin improving model generalizability.

Deep learning models have achieved exceptional predictive performance in a wide variety of tasks, ranging from computer vision, natural language processing, to graph mining. Many businesses and organizations across diverse domains are now building large-scale applications based on deep learning. However, there are growing concerns, regarding the fairness, security, and trustworthiness of these models, largely due to the opaque nature of their decision processes. Recently, there has been an increasing interest in explainable deep learning that aims to reduce the opacity of a model by explaining its behavior, its predictions, or both, thus building trust between human and complex deep learning models. A collection of explanation methods have been proposed in recent years that address the problem of low explainability and opaqueness of models. In this tutorial, we introduce recent explanation methods from a data perspective, targeting models that process image data, text data, and graph data, respectively. We will compare their strengths and limitations, and offer real-world applications.

Training graph neural networks (GNNs) with good generalizability on large-scale graphs is a challenging problem. Existing methods mainly divide the input graph into multiple subgraphs and train them in different batches to improve training scalability. However, the local batches obtained by such a strategy could contain topological bias compared with the complete graph structure. It has been studied that the topological bias results in more significant gaps between training and testing performances, or worse generalization robustness. A straightforward solution is to utilize contrastive learning, and train node embeddings to be robust and invariant among the augmented imperfect graphs. However, most of the existing work are inefficient by contrasting extensive node pairs at the large-scale graph. With random data augmentation, they may deteriorate the embedding process by transforming well-sampled batches into meaningless graph structures.

Graph Neural Networks (GNNs) are playing increasingly important roles in critical decision-making scenarios due to their exceptional performance and end-to-end design. However, concerns have been raised that GNNs could make biased decisions against underprivileged groups or individuals. To remedy this issue, researchers have proposed various fairness notions including individual fairness that gives similar predictions to similar individuals. However, existing methods in individual fairness rely on Lipschitz condition: they only optimize overall individual fairness and disregard equality of individual fairness between groups. This leads to drastically different levels of individual fairness among groups. We tackle this problem by proposing a novel GNN framework GUIDE to achieve group equality informed individual fairness in GNNs. We aim to not only achieve individual fairness but also equalize the levels of individual fairness among groups. Specifically, our framework operates on the similarity matrix of individuals to learn personalized attention to achieve individual fairness without group level disparity. Comprehensive experiments on real-world datasets demonstrate that GUIDE obtains good balance of group equality informed individual fairness and model utility. The open-source implementation of GUIDE can be found here: https://github.com/mikesong724/GUIDE.

With the advancement of data science and AI, more and more powerful and accurate recommender systems (RSs) have been developed. They provide recommendation services in various areas, including shopping, eating, travelling and entertainment. RSs have achieved a great success and benefted the society. However, most of the research on RS has focused on the improvement of the recommendation accuracy, while ignoring other important qualities, such as trustworthiness (robustness, fairness, explainability, privacy and security) and social impact (influence on users' recognition and behaviours) of the recommendations. These are important aspects and cannot be overlooked since they measure properties that determine whether the recommendation service is reliable, trustworthy and benefcial to individual users and society. In this work, responsible recommendations refer to trustworthy recommendation techniques and positive-social-impact recommendation results.

Network anomaly detection is a crucial task since a few anomalies can cause huge losses. Semi-supervised anomaly detection methods can effectively leverage a small number of labels as prior knowledge to enhance detection accuracy. But in real-world scenarios, novel types of anomalies (i.e., unseen anomalies) usually exist on networks which may present different characteristics with the seen anomalies and are hard to be identified by prior semi-supervised anomaly detection methods. In this paper, we propose the novel problem of unseen network anomaly detection that aims to identify both seen and unseen anomalies to eliminate potential dangers. Accordingly, we propose a method called Multi-hypersphere Graph Learning (MHGL) to effectively leverage existing labels by learning fine-grained normal patterns to discriminate anomalies. Experiments demonstrate that MHGL outperforms state-of-the-art methods significantly.

Graph Neural Networks (GNNs) have shown superior performance in analyzing attributed networks in various web-based applications such as social recommendation and web search. Nevertheless, in high-stake decision-making scenarios such as online fraud detection, there is an increasing societal concern that GNNs could make discriminatory decisions towards certain demographic groups. Despite recent explorations on fair GNNs, these works are tailored for a specific GNN model. However, myriads of GNN variants have been proposed for different applications, and it is costly to fine-tune existing debiasing algorithms for each specific GNN architecture. Different from existing works that debias GNN models, we aim to debias the input attributed network to achieve fairer GNNs through feeding GNNs with less biased data. Specifically, we propose novel definitions and metrics to measure the bias in an attributed network, which leads to the optimization objective to mitigate bias. We then develop a framework EDITS to mitigate the bias in attributed networks while maintaining the performance of GNNs in downstream tasks. EDITS works in a model-agnostic manner, i.e., it is independent of any specific GNN. Experiments demonstrate the validity of the proposed bias metrics and the superiority of EDITS on both bias mitigation and utility maintenance. Open-source implementation: https://github.com/yushundong/EDITS.

Learning discriminative node representations benefits various downstream tasks in graph analysis such as community detection and node classification. Existing graph representation learning methods (e.g., based on random walk and contrastive learning) are limited to maximizing the local similarity of connected nodes. Such pair-wise learning schemes could fail to capture the global distribution of representations, since it has no explicit constraints on the global geometric properties of representation space. To this end, we propose Geometric Graph Representation Learning (G2R) to learn node representations in an unsupervised manner via maximizing rate reduction. In this way, G2R maps nodes in distinct groups (implicitly stored in the adjacency matrix) into different subspaces, while each subspace is compact and different subspaces are dispersedly distributed. G2R adopts a graph neural network as the encoder and maximizes the rate reduction with the adjacency matrix. Furthermore, we theoretically and empirically demonstrate that rate reduction maximization is equivalent to maximizing the principal angles between different subspaces. Experiments on real-world datasets show that G2R outperforms various baselines on node classification and community detection tasks.

Recent methods in sequential recommendation focus on learning an overall embedding vector from a user's behavior sequence for the next-item recommendation. However, from empirical analysis, we discovered that a user's behavior sequence often contains multiple conceptually distinct items, while a unified embedding vector is primarily affected by one's most recent frequent actions. Thus, it may fail to infer the next preferred item if conceptually similar items are not dominant in recent interactions. To this end, an alternative solution is to represent each user with multiple embedding vectors encoding different aspects of the user's intentions. Nevertheless, recent work on multi-interest embedding usually considers a small number of concepts discovered via clustering, which may not be comparable to the large pool of item categories in real systems. It is a non-trivial task to effectively model a large number of diverse conceptual prototypes, as items are often not conceptually well clustered in fine granularity. Besides, an individual usually interacts with only a sparse set of concepts. In light of this, we propose a novel Sparse Interest NEtwork (SINE) for sequential recommendation. Our sparse-interest module can adaptively infer a sparse set of concepts for each user from the large concept pool and output multiple embeddings accordingly. Given multiple interest embeddings, we develop an interest aggregation module to actively predict the user's current intention and then use it to explicitly model multiple interests for next-item prediction. Empirical results on several public benchmark datasets and one large-scale industrial dataset demonstrate that SINE can achieve substantial improvement over state-of-the-art methods.

Recommender systems play a fundamental role in web applications in filtering massive information and matching user interests. While many efforts have been devoted to developing more effective models in various scenarios, the exploration on the explainability of recommender systems is running behind. Explanations could help improve user experience and discover system defects. In this paper, after formally introducing the elements that are related to model explainability, we propose a novel explainable recommendation model through improving the transparency of the representation learning process. Specifically, to overcome the representation entangling problem in traditional models, we revise traditional graph convolution to discriminate information from different layers. Also, each representation vector is factorized into several segments, where each segment relates to one semantic aspect in data. Different from previous work, in our model, factor discovery and representation learning are simultaneously conducted, and we are able to handle extra attribute information and knowledge. In this way, the proposed model can learn interpretable and meaningful representations for users and items. Unlike traditional methods that need to make a trade-off between explainability and effectiveness, the performance of our proposed explainable model is not negatively affected after considering explainability. Finally, comprehensive experiments are conducted to validate the performance of our model as well as explanation faithfulness.

With the widespread use of deep neural networks (DNNs) in high-stake applications, the security problem of the DNN models has received extensive attention. In this paper, we investigate a specific security problem called trojan attack, which aims to attack deployed DNN systems relying on the hidden trigger patterns inserted by malicious hackers. We propose a training-free attack approach which is different from previous work, in which trojaned behaviors are injected by retraining model on a poisoned dataset. Specifically, we do not change parameters in the original model but insert a tiny trojan module (TrojanNet) into the target model. The infected model with a malicious trojan can misclassify inputs into a target label when the inputs are stamped with the special trigger. The proposed TrojanNet has several nice properties including (1) it activates by tiny trigger patterns and keeps silent for other signals, (2) it is model-agnostic and could be injected into most DNNs, dramatically expanding its attack scenarios, and (3) the training-free mechanism saves massive training efforts comparing to conventional trojan attack methods. The experimental results show that TrojanNet can inject the trojan into all labels simultaneously (all-label trojan attack) and achieves 100% attack success rate without affecting model accuracy on original tasks. Experimental analysis further demonstrates that state-of-the-art trojan detection algorithms fail to detect TrojanNet attack. The code is available at https://github.com/trx14/TrojanNet.

Deep neural network (DNN) has become an effective computational tool because of its superior performance in practice. However, the generalization of DNN still largely depends on the training data, no matter in quantity or quality. In this paper, we propose a knowledge instillation framework, named NeuKI, for feed-forward DNN, aiming to enhance learning performance with the aid of knowledge. This task is particularly challenging due to the complicated nature of knowledge and numerous variants of DNN architectures. To bridge the gap, we construct a separate knowledge-DNN faithfully encoding the instilled knowledge for joint training. The core idea is to regularize the training of target-DNN with the constructed knowledge-DNN, so that the instilled knowledge can guide the model training. The proposed NeuKI is demonstrated to be applicable to both knowledge rules and constraints, where rules are encoded by structure and constraints are handled by loss. Experiments are conducted on several real-world datasets from different domains, and the results demonstrate the effectiveness of NeuKI in improving learning performance, as well as relevant data efficiency and model interpretability.

Recommender systems in industry generally include two stages: recall and ranking. Recall refers to efficiently identify hundreds of candidate items that user may interest in from a large volume of item corpus, while the latter aims to output a precise ranking list using complex ranking models. Recently, graph representation learning has attracted much attention in supporting high quality candidate search at scale. Despite its effectiveness in learning embedding vectors for objects in the user-item interaction network, the computational costs to infer users’ preferences in continuous embedding space are tremendous. In this work, we investigate the problem of hashing with graph neural networks (GNNs) for high quality retrieval, and propose a simple yet effective discrete representation learning framework to jointly learn continuous and discrete codes. Specifically, a deep hashing with GNNs (HashGNN) is presented, which consists of two components, a GNN encoder for learning node representations, and a hash layer for encoding representations to hash codes. The whole architecture is trained end-to-end by jointly optimizing two losses, i.e., reconstruction loss from reconstructing observed links, and ranking loss from preserving the relative ordering of hash codes. A novel discrete optimization strategy based on straight through estimator (STE) with guidance is proposed. The principal idea is to avoid gradient magnification in back-propagation of STE with continuous embedding guidance, in which we begin from learning an easier network that mimic the continuous embedding and let it evolve during the training until it finally goes back to STE. Comprehensive experiments over three publicly available and one real-world Alibaba company datasets demonstrate that our model not only can achieve comparable performance compared with its continuous counterpart but also runs multiple times faster during inference.

Recent explainability related studies have shown that state-of-the-art DNNs do not always adopt correct evidences to make decisions. It not only hampers their generalization but also makes them less likely to be trusted by end-users. In pursuit of developing more credible DNNs, in this paper we propose CREX, which encourages DNN models to focus more on evidences that actually matter for the task at hand, and to avoid overfitting to data-dependent bias and artifacts. Specifically, CREX regularizes the training process of DNNs with rationales, i.e., a subset of features highlighted by domain experts as justifications for predictions, to enforce DNNs to generate local explanations that conform with expert rationales. Even when rationales are not available, CREX still could be useful by requiring the generated explanations to be sparse. Experimental results on two text classification datasets demonstrate the increased credibility of DNNs trained with CREX. Comprehensive analysis further shows that while CREX does not always improve prediction accuracy on the held-out test set, it significantly increases DNN accuracy on new and previously unseen data beyond test set, highlighting the advantage of the increased credibility.

Networks have been widely used as the data structure for abstracting real-world systems as well as organizing the relations among entities. Network embedding models are powerful tools in mapping nodes in a network into continuous vector-space representations in order to facilitate subsequent tasks such as classification and link prediction. Existing network embedding models comprehensively integrate all information of each node, such as links and attributes, towards a single embedding vector to represent the node's general role in the network. However, a real-world entity could be multifaceted, where it connects to different neighborhoods due to different motives or self-characteristics that are not necessarily correlated. For example, in a movie recommender system, a user may love comedies or horror movies simultaneously, but it is not likely that these two types of movies are mutually close in the embedding space, nor the user embedding vector could be sufficiently close to them at the same time. In this paper, we propose a polysemous embedding approach for modeling multiple facets of nodes, as motivated by the phenomenon of word polysemy in language modeling. Each facet of a node is mapped to an embedding vector, while we also maintain an association degree between each pair of node and facet. The proposed method is adaptive to various existing embedding models, without significantly complicating the optimization process. We also discuss how to engage embedding vectors of different facets for inference tasks including classification and link prediction. Experiments on real-world datasets help comprehensively evaluate the performance of the proposed method.

Representation learning models map data instances into a low-dimensional vector space, thus facilitating the deployment of subsequent models such as classification and clustering models, or the implementation of downstream applications such as recommendation and anomaly detection. However, the outcome of representation learning is difficult to be directly understood by users, since each dimension of the latent space may not have any specific meaning. Understanding representation learning could be beneficial to many applications. For example, in recommender systems, knowing why a user instance is mapped to a certain position in the latent space may unveil the user's interests and profile. In this paper, we propose an interpretation framework to understand and describe how representation vectors distribute in the latent space. Specifically, we design a coding scheme to transform representation instances into spatial codes to indicate their locations in the latent space. Following that, a multimodal autoencoder is built for generating the description of a representation instance given its spatial codes. The coding scheme enables indication of position with different granularity. The incorporation of autoencoder makes the framework capable of dealing with different types of data. Several metrics are designed to evaluate interpretation results. Experiments under various application scenarios and different representation learning models are conducted to demonstrate the flexibility and effectiveness of the proposed framework.

RNN models have achieved the state-of-the-art performance in a wide range of text mining tasks. However, these models are often regarded as black-boxes and are criticized due to the lack of interpretability. In this paper, we enhance the interpretability of RNNs by providing interpretable rationales for RNN predictions. Nevertheless, interpreting RNNs is a challenging problem. Firstly, unlike existing methods that rely on local approximation, we aim to provide rationales that are more faithful to the decision making process of RNN models. Secondly, a flexible interpretation method should be able to assign contribution scores to text segments of varying lengths, instead of only to individual words. To tackle these challenges, we propose a novel attribution method, called REAT, to provide interpretations to RNN predictions. REAT decomposes the final prediction of a RNN into additive contribution of each word in the input text. This additive decomposition enables REAT to further obtain phrase-level attribution scores. In addition, REAT is generally applicable to various RNN architectures, including GRU, LSTM and their bidirectional versions. Experimental results demonstrate the faithfulness and interpretability of the proposed attribution method. Comprehensive analysis shows that our attribution method could unveil the useful linguistic knowledge captured by RNNs. Some analysis further demonstrates our method could be utilized as a debugging tool to examine the vulnerability and failure reasons of RNNs, which may lead to several promising future directions to promote generalization ability of RNNs.

Despite the wide application in recent years, most recommender systems are not capable of providing interpretations together with recommendation results, which impedes both deployers and customers from understanding or trusting the results. Recent advances in recommendation models, such as deep learning models, usually involve extracting latent representations of users and items. However, the representation space is not directly comprehensible since each dimension usually does not have any specific meaning. In addition, recommender systems incorporate various sources of information, such as user behaviors, item information, and other side content information. Properly organizing different types of information, as well as effectively selecting important information for interpretation, is challenging and has not been fully tackled by conventional interpretation methods. In this paper, we propose a post-hoc method called Sorted Explanation Paths (SEP) to interpret recommendation results. Specifically, we first build a unified heterogeneous information network to incorporate multiple types of objects and relations based on representations from the recommender system and information from the dataset. Then, we search for explanation paths between given recommendation pairs, and use the set of simple paths to construct semantic explanations. Next, three heuristic metrics, i.e., credibility, readability and diversity, are designed to measure the validity of each explanation path, and to sort all the paths comprehensively. The top-ranked explanation paths are selected as the final interpretation. After that, practical issues on computation and efficiency of the proposed SEP method are also handled by corresponding approaches. Finally, we conduct experiments on three real-world benchmark datasets, and demonstrate the applicability and effectiveness of the proposed SEP method.

While deep neural networks (DNN) have become an effective computational tool, the prediction results are often criticized by the lack of interpretability, which is essential in many real-world applications such as health informatics. Existing attempts based on local interpretations aim to identify relevant features contributing the most to the prediction of DNN by monitoring the neighborhood of a given input. They usually simply ignore the intermediate layers of the DNN that might contain rich information for interpretation. To bridge the gap, in this paper, we propose to investigate a guided feature inversion framework for taking advantage of the deep architectures towards effective interpretation. The proposed framework not only determines the contribution of each feature in the input but also provides insights into the decision-making process of DNN models. By further interacting with the neuron of the target category at the output layer of the DNN, we enforce the interpretation result to be class-discriminative. We apply the proposed interpretation model to different CNN architectures to provide explanations for image data and conduct extensive experiments on ImageNet and PASCAL VOC07 datasets. The interpretation results demonstrate the effectiveness of our proposed framework in providing class-discriminative interpretation for DNN-based prediction.

Network embedding has been increasingly used in many network analytics applications to generate low-dimensional vector representations, so that many off-the-shelf models can be applied to solve a wide variety of data mining tasks. However, similar to many other machine learning methods, network embedding results remain hard to be understood by users. Each dimension in the embedding space usually does not have any specific meaning, thus it is difficult to comprehend how the embedding instances are distributed in the reconstructed space. In addition, heterogeneous content information may be incorporated into network embedding, so it is challenging to specify which source of information is effective in generating the embedding results. In this paper, we investigate the interpretation of network embedding, aiming to understand how instances are distributed in embedding space, as well as explore the factors that lead to the embedding results. We resort to the post-hoc interpretation scheme, so that our approach can be applied to different types of embedding methods. Specifically, the interpretation of network embedding is presented in the form of a taxonomy. Effective objectives and corresponding algorithms are developed towards building the taxonomy. We also design several metrics to evaluate interpretation results. Experiments on real-world datasets from different domains demonstrate that, by comparing with the state-of-the-art alternatives, our approach produces effective and meaningful interpretation to embedding results.

Machine learning (ML) systems have been increasingly applied in web security applications such as spammer detection, malware detection and fraud detection. These applications have an intrinsic adversarial nature where intelligent attackers can adaptively change their behaviors to avoid being detected by the deployed detectors. Existing efforts against adversaries are usually limited by the type of applied ML models or the specific applications such as image classification. Additionally, the working mechanisms of ML models usually cannot be well understood by users, which in turn impede them from understanding the vulnerabilities of models nor improving their robustness. To bridge the gap, in this paper, we propose to investigate whether model interpretation could potentially help adversarial detection. Specifically, we develop a novel adversary-resistant detection framework by utilizing the interpretation of ML models. The interpretation process explains the mechanism of how the target ML model makes prediction for a given instance, thus providing more insights for crafting adversarial samples. The robustness of detectors is then improved through adversarial training with the adversarial samples. A data-driven method is also developed to empirically estimate costs of adversaries in feature manipulation. Our approach is model-agnostic and can be applied to various types of classification models. Our experimental results on two real-world datasets demonstrate the effectiveness of interpretation-based attacks and how estimated feature manipulation cost would affect the behavior of adversaries.