Workflow automation promises substantial productivity gains in everyday document-related tasks. While prior agentic systems can execute isolated instructions, they struggle with automating multi-step, session-level workflows due to limited control over the operational process. To this end, we introduce AutoDW, a novel execution framework that enables stepwise, rollback-enabled operation orchestration. AutoDW incrementally plans API actions conditioned on user instructions, intent-filtered API candidates, and the evolving states of the document. It further employs robust rollback mechanisms at both the argument and API levels, enabling dynamic correction and fault tolerance. These designs together ensure that the execution trajectory of AutoDW remains aligned with user intent and document context across long-horizon workflows. To assess its effectiveness, we construct a comprehensive benchmark of 250 sessions and 1,708 human-annotated instructions, reflecting realistic document processing scenarios with interdependent instructions. AutoDW achieves 90% and 62% completion rates on instruction- and session-level tasks, respectively, outperforming strong baselines by 40% and 76%. Moreover, AutoDW also remains robust for the decision of backbone LLMs and on tasks with varying difficulty.

Despite the rapid advancements of electrocardiogram (ECG) signal diagnosis and analysis methods through deep learning, two major hurdles still limit their clinical adoption: the lack of versatility in processing ECG signals with diverse configurations, and the inadequate detection of risk signals due to sample imbalances. Addressing these challenges, we introduceVersAtile andRisk-Sensitive cardiac diagnosis (VARS), an innovative approach that employs a graph-based representation to uniformly model heterogeneous ECG signals. VARS stands out by transforming ECG signals into versatile graph structures that capture critical diagnostic features, irrespective of signal diversity in the lead count, sampling frequency, and duration. This graph-centric formulation also enhances diagnostic sensitivity, enabling precise localization and identification of abnormal ECG patterns that often elude standard analysis methods. To facilitate representation transformation, our approach integrates denoising reconstruction with contrastive learning to preserve raw ECG information while highlighting pathognomonic patterns. We rigorously evaluate the efficacy of VARS on three distinct ECG datasets, encompassing a range of structural variations. The results demonstrate that VARS not only consistently surpasses existing state-of-the-art models across all these datasets but also exhibits substantial improvement in identifying risk signals. Additionally, VARS offers interpretability by pinpointing the exact waveforms that lead to specific model outputs, thereby assisting clinicians in making informed decisions. These findings suggest that our VARS will likely emerge as an invaluable tool for comprehensive cardiac health assessment.

Zhuoyun Du, LujieZheng LujieZheng, Renjun Hu, Yuyang Xu, Xiawei Li, Ying Sun, Wei Chen, Jian Wu, Haolei Cai, Haochao Ying. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

The increasing popularity of online food blogs and food ordering services has made personalized recipe recommendation a vital aspect of our emotional well-being. However, existing solutions, mainly based on graph neural networks, still face significant challenges, such as (a) focusing on exploiting the user-recipe interactions while neglecting other crucial pairwise and high-order relationships, and (b) failing to explicitly distinguish the distinct factors, e.g., hedonic and healthy, that influence recipe selection. To address these issues, we propose a progressively-passing-then-disentangling approach named P2D. Our approach utilizes a three-stage progressive message-passing mechanism for better representation learning. Specifically, we incorporate the extra pairwise relationships between recipes and nutrients, ingredients, and visual contents to create fine-grained and multimodal recipe representations. We next refine these representations via message passing between high-order recipe relationships to learn people's shared food preferences. Based on them, we could derive comprehensive user representations, which are subsequently transformed into disentangled forms that correspond to various decision factors through contrastive and mutual information regularization. Experimental results demonstrate both the superiority and the rationality of our method: (a) P2D outperforms the state-of-the-art recipe recommendation methods by a large margin under various metrics, (b) ablation studies confirm the positive impact of each of its components, and (c) our visualization analysis empirically supports the advantage of explicitly disentangling decision factors.

Until recently, the question of the effective inductive bias of deep models on tabular data has remained unanswered. This paper investigates the hypothesis that arithmetic feature interaction is necessary for deep tabular learning. To test this point, we create a synthetic tabular dataset with a mild feature interaction assumption and examine a modified transformer architecture enabling arithmetical feature interactions, referred to as AMFormer. Results show that AMFormer outperforms strong counterparts in fine-grained tabular data modeling, data efficiency in training, and generalization. This is attributed to its parallel additive and multiplicative attention operators and prompt-based optimization, which facilitate the separation of tabular samples in an extended space with arithmetically-engineered features. Our extensive experiments on real-world data also validate the consistent effectiveness, efficiency, and rationale of AMFormer, suggesting it has established a strong inductive bias for deep learning on tabular data. Code is available at https://github.com/aigc-apps/AMFormer.

Generative query rewrite generates reconstructed query rewrites using the conversation history while rely heavily on gold rewrite pairs that are expensive to obtain. Recently, few-shot learning is gaining increasing popularity for this task, whereas these methods are sensitive to the inherent noise due to limited data size. Besides, both attempts face performance degradation when there exists language style shift between training and testing cases. To this end, we study low-resource generative conversational query rewrite that is robust to both noise and language style shift. The core idea is to utilize massive unlabeled data to make further improvements via a contrastive co-training paradigm. Specifically, we co-train two dual models (namely Rewriter and Simplifier) such that each of them provides extra guidance through pseudo-labeling for enhancing the other in an iterative manner. We also leverage contrastive learning with data augmentation, which enables our model pay more attention on the truly valuable information than the noise. Extensive experiments demonstrate the superiority of our model under both few-shot and zero-shot scenarios. We also verify the better generalization ability of our model when encountering language style shift.

Expert-designed close-ended benchmarks are indispensable in assessing the knowledge capacity of large language models (LLMs). Despite their widespread use, concerns have mounted regarding their reliability due to limited test scenarios and an unavoidable risk of data contamination. To rectify this, we present PertEval, a toolkit devised for in-depth probing of LLMs' knowledge capacity through knowledge-invariant perturbations. These perturbations employ human-like restatement techniques to generate on-the-fly test samples from static benchmarks, meticulously retaining knowledge-critical content while altering irrelevant details. Our toolkit further includes a suite of response consistency analyses that compare performance on raw vs. perturbed test sets to precisely assess LLMs' genuine knowledge capacity. Six representative LLMs are re-evaluated using PertEval. Results reveal significantly inflated performance of the LLMs on raw benchmarks, including an absolute 25.8% overestimation for GPT-4. Additionally, through a nuanced response pattern analysis, we discover that PertEval retains LLMs' uncertainty to specious knowledge, and reveals their potential rote memorization to correct options which leads to overestimated performance. We also find that the detailed response consistency analyses by PertEval could illuminate various weaknesses in existing LLMs' knowledge mastery and guide the development of refinement. Our findings provide insights for advancing more robust and genuinely knowledgeable LLMs. Our code is available at https://github.com/aigc-apps/PertEval.

The advent of third-generation semiconductors brings the switching frequency into a much higher level, which enables high-frequency transformer with a smaller footprint in power electronics. Due to its inherent ability of voltage amplification, the transformer-based voltage-multiplier becomes prevalent in step-up converters which are the inevitable parts in the usage of low-voltage renewable resources. Though its feasibility has been verified in many researches, an uncertain error in voltage conversion ratio exists between the reality and ideal, which leaves a theoretical gap. Accordingly, the key affecting characteristics, the reverse recovery process of the leakage inductance, is considered and analyzed in this paper. The power transmission mechanism is clarified, and the root-mean-square error of the gain can be reduced from 24.31% to 3.13%, and the maximum output power together with power effect on the gain are obtained. With the conclusion of this paper, parameters of power supply can be optimized at the beginning of design to satisfy the demand of maximum output power and gain over a wide power range, which can save the unnecessary trial-and-error during the case-by-case design.

Image ordinal regression has been mainly studied along the line of exploiting the order of categories. However, the issues of class imbalance and category overlap that are very common in ordinal regression were largely overlooked. As a result, the performance on minority categories is often unsatisfactory. In this paper, we propose a novel framework called CIG based on controllable image generation to directly tackle these two issues. Our main idea is to generate extra training samples with specific labels near category boundaries, and the sample generation is biased toward the less-represented categories. To achieve controllable image generation, we seek to separate structural and categorical information of images based on structural similarity, categorical similarity, and reconstruction constraints. We evaluate the effectiveness of our new CIG approach in three different image ordinal regression scenarios. The results demonstrate that CIG can be flexibly integrated with off-the-shelf image encoders or ordinal regression models to achieve improvement, and further, the improvement is more significant for minority categories.

Graph Neural Networks (GNNs) have become widely-used models for semi-supervised learning. However, the robustness of GNNs in the presence of label noise remains a largely under-explored problem. In this paper, we consider an important yet challenging scenario where labels on nodes of graphs are not only noisy but also scarce. In this scenario, the performance of GNNs is prone to degrade due to label noise propagation and insufficient learning. To address these issues, we propose a novel RTGNN (Robust Training of Graph Neural Networks via Noise Governance) framework that achieves better robustness by learning to explicitly govern label noise. More specifically, we introduce self-reinforcement and consistency regularization as supplemental supervision. The self-reinforcement supervision is inspired by the memorization effects of deep neural networks and aims to correct noisy labels. Further, the consistency regularization prevents GNNs from overfitting to noisy labels via mimicry loss in both the inter-view and intra-view perspectives. To leverage such supervisions, we divide labels into clean and noisy types, rectify inaccurate labels, and further generate pseudo-labels on unlabeled nodes. Supervision for nodes with different types of labels is then chosen adaptively. This enables sufficient learning from clean labels while limiting the impact of noisy ones. We conduct extensive experiments to evaluate the effectiveness of our RTGNN framework, and the results validate its consistent superior performance over state-of-the-art methods with two types of label noises and various noise rates.

Competition between Points of Interest (POIs) refers to the situation in which two POIs directly or indirectly provide similar services to secure businesses. A large portion of prior studies on competition analysis focuses on mining textual data, e.g., news articles and social comments. However, the increasing availability of human mobility and mobile query data enables a new paradigm for analyzing the competitive relationships among POIs, which remains largely unexplored. To this end, in this paper, we attempt to mine large-scale online map search query data for better understanding POI competitive relationships. Based on a co-query POI graph built from the map search query data, we develop a novel neural graphlet-based prediction framework to predict the competitive relationships among POIs. A unique perspective of our model is to infer latent POI competitive relationships by integrating multiple distinct factors, e.g., graphlet structure, geographical distance, and regional features, reflected in map search query data and POI data. Finally, we conduct extensive experiments on real-world datasets to demonstrate the effectiveness of the proposed framework, and show that our framework outperforms all baselines with a significant margin in all evaluation metrics.

Understanding human mobility patterns at the point-of-interest (POI) scale plays an important role in enhancing business intelligence in mobile environments. While large efforts have been made in this direction, most studies simply utilize POI check-ins to mine the concerned mobility patterns, the effectiveness of which is usually hindered due to data sparsity. To obtain better POI-based human mobility for mining, in this paper, we strive to directly annotate the POIs associated with raw user-generated mobility records. We propose a neural context fusion approach which integrates various context factors in people's POI-visiting behaviors. Our approach evaluates the preference and transition factors via representation learning. Notably, we incorporate an attention mechanism to deal with the randomized transitions in raw mobility. The domain knowledge factors, i.e., distance, time and popularity, remain effective and our approach further includes them from a data-driven perspective. Factors are automatically fused with a feed-forward neural network. Furthermore, we exploit a multi-head architecture to enhance the model expressiveness. Using two real-life data sets, we conduct our experimental study and find that our approach consistently outperforms the state-of-the-art baselines by at least 32 percent in accuracy. Besides, we demonstrate the utility of the obtained POI-based human mobility with a POI recommendation example.

Event-based corporate profiling aims to assess the evolving operational status of the corresponding corporate from its event sequence. Existing studies on corporate profiling have partially addressed the problem via (i) case-by-case empirical analysis by leveraging traditional financial methods, or (ii) the automatic profile inference by reformulating the problem into a supervised learning task. However, both approaches heavily rely on domain knowledge and are labor-intensive. More importantly, the task-specific nature of both approaches prevents the obtained corporate profiles from being applied to diversified downstream applications. To this end, in this paper, we propose a Self-Supervised Prototype Representation Learning (SePaL) framework for dynamic corporate profiling. By exploiting the topological information of an event graph and exploring self-supervised learning techniques, SePaL can obtain unified corporate representations that are robust to event noises and can be easily fine-tuned to benefit various down-stream applications with only a few annotated data. Specifically, we first infer the initial cluster distribution of noise-resistant event prototypes based on latent representations of events. Then, we construct four permutation-invariant self-supervision signals to guide the representation learning of the event prototype. In terms of applications, we exploit the learned time-evolving corporate representations for both stock price spike prediction and corporate default risk evaluation. Experimental results on two real-world corporate event datasets demonstrate the effectiveness of SePaL for these two applications.

Query and Point-of-Interest (POI) matching, aiming at recommending the most relevant POIs from partial query keywords, has become one of the most essential functions in online navigation and ride-hailing applications. Existing methods for query-POI matching, such as Google Maps and Uber, have a natural focus on measuring the static semantic similarity between contextual information of queries and geographical information of POIs. However, it remains challenging for dynamic and personalized online query-POI matching because of the non-stationary and situational context-dependent query-POI relevance. Moreover, the large volume of online queries requires an adaptive and incremental model training strategy that is efficient and scalable in the online scenario. To this end, in this paper, we propose an Incremental Spatio-Temporal Graph Learning (IncreSTGL) framework for intelligent online query-POI matching. Specifically, we first model dynamic query-POI interactions as microscopic and macroscopic graphs. Then, we propose an incremental graph representation learning module to refine and update query-POI interaction graphs in an online incremental fashion, which includes: (i) a contextual graph attention operation quantifying query-POI correlation based on historical queries under dynamic situational context, (ii) a graph discrimination operation capturing the sequential query-POI relevance drift from a holistic view of personalized preference and social homophily, and (iii) a multi-level temporal attention operation summarizing the temporal variations of query-POI interaction graphs for subsequent query-POI matching. Finally, we introduce a lightweight semantic matching module for online query-POI similarity measurement. To demonstrate the effectiveness and efficiency of the proposed algorithm, we conduct extensive experiments on two real-world datasets collected from a leading online navigation and map service provider in China.

While Point-of-Interest (POI) recommendation has been a popular topic of study for some time, little progress has been made for understanding why and how people make their decisions for the selection of POIs. To this end, in this paper, we propose a user decision profiling framework, named PROUD, which can identify the key factors in people's decisions on choosing POIs. Specifically, we treat each user decision as a set of factors and provide a method for learning factor embeddings. A unique perspective of our approach is to identify key factors, while preserving decision structures seamlessly, via a novel scalar projection maximization objective. Exactly solving the objective is non-trivial due to a sparsity constraint. To address this, our PROUD adopts a self projection attention and an L2 regularized sparse activation to directly estimate the likelihood of each factor to be a key factor. Finally, extensive experiments on real-world data validate the advantage of PROUD in preserving user decision structures. Also, our case study indicates that the identified key decision factors can help us to provide more interpretable recommendations and analyses.

Scholarly search systems greatly aid the deep understanding of scholarly data and facilitate the research activities of scholars for scientific studies. Though a number of such systems have been developed, most of them either support rankings of limited search of entities or provide only basic ranking metrics. These existing systems also mainly adopt RDBMSs as their storage such that the linked feature of scholarly data is not fully exploited. In this study, we design and develop a novel scholarly search system Athena. (1) It supports four types of scholarly entity searches: articles, authors, venues and affiliations, and is equipped with five ranking metrics, including three traditional metrics and two comprehensive importance ranking metrics. (2) It also provides profiling of scholarly entities. (3) It further utilizes a graph storage to directly leverage the linked feature for speeding up the processing of complex queries. We demonstrate the advantages of Athena at scholarly search, profiling, graph storage and ranking quality.

Human mobility annotation aims to assign mobility records the corresponding visiting Point-of-Interests (POIs). It is one of the most fundamental problems for understanding human mobile behaviors. In literature, many efforts have been devoted to annotating mobility records in a pointwise or trajectory-wise manner. However, the user preference factor is not fully explored and, worse still, the mobile peer influence factor has never been integrated. To this end, in this article, we propose a novel framework, named JEPPI, to jointly exploit user preference and mobile peer influence to tackle the problem. In our JEPPI, we first unify the two distinct factors in a behavior-driven user-POI graph. This graph enables us to model user preference with user-POI visiting relationships, and model two types of mobile peer influence with co-location and co-visiting peer relationships, respectively. Moreover, we devise an equivalence-emphasizing metric to reduce redundancy in the second-order co-visiting peer influence. In addition, a mutual augmentation learning approach is proposed to preserve the latent structures of various factors exploited. Notably, our learning approach preserves all factors in a shared representation space such that user preference is learned with mobile peer influence being considered at the same time, and vice versa. In this way, the different factors are mutually augmented and semantically integrated to enhance human mobility annotation. Finally, using two large-scale real-world datasets, we conduct extensive experiments to demonstrate the superiority of our approach compared with the state-of-the-art annotation methods.

Dense subgraph discovery has proven useful in various applications of temporal networks. We focus on a special class of temporal networks whose nodes and edges are kept fixed, but edge weights regularly vary with timestamps. However, finding dense subgraphs in temporal networks is non-trivial, and its state of the art solution uses a filter-and-verification framework that is not scalable on large temporal networks. In this study, we propose a highly efficient approach to finding dense subgraphs in large temporal networks with T timestamps. (1) We first develop a statistics-driven approach that employs hidden statistics to identifying k time intervals, instead of T(T + 1)/2 ones (k is typically much smallerthan T), which strikes a balance between quality and efficiency. (2) After proving that the problem has no constant factor approximation algorithms, we design better heuristic algorithms to attack the problem, by connecting finding dense subgraphs with a variant of the Prize Collecting Steiner Tree problem. (3) Finally, we have conducted an extensive experimental study to verify that our approach is both effective and efficient.

Multi-modal transportation recommendation has a goal of recommending a travel plan which considers various transportation modes, such as walking, cycling, automobile, and public transit, and how to connect among these modes. The successful development of multi-modal transportation recommendation systems can help to satisfy the diversified needs of travelers and improve the efficiency of transport networks. However, existing transport recommender systems mainly focus on unimodal transport planning. To this end, in this paper, we propose a joint representation learning framework for multi-modal transportation recommendation based on a carefully-constructed multi-modal transportation graph. Specifically, we first extract a multi-modal transportation graph from large-scale map query data to describe the concurrency of users, Origin-Destination (OD) pairs, and transport modes. Then, we provide effective solutions for the optimization problem and develop an anchor embedding for transport modes to initialize the embeddings of transport modes. Moreover, we infer user relevance and OD pair relevance, and incorporate them to regularize the representation learning. Finally, we exploit the learned representations for online multimodal transportation recommendations. Indeed, our method has been deployed into one of the largest navigation Apps to serve hundreds of millions of users, and extensive experimental results with real-world map query data demonstrate the enhanced performance of the proposed method for multimodal transportation recommendations.

Tags of a Point of Interest (POI) can facilitate location-based services from many aspects like location search and place recommendation. However, many POI tags are often incomplete or imprecise, which may lead to performance degradation of tag-dependent applications. In this paper, we study the POI tag refinement problem which aims to automatically fill in the missing tags as well as correct noisy tags for POIs. We propose a tri-adaptive collaborative learning framework to search for an optimal POI-tag score matrix. The framework integrates three components to collaboratively (i) model the similarity matching between POI and tag, (ii) recover the POI-tag pattern via matrix factorization and (iii) learn to infer the most possible tags by maximum likelihood estimation. We devise an adaptively joint training process to optimize the model and regularize each component simultaneously. And the final refinement results are the consensus of multiple views from different components. We also discuss how to utilize various data sources to construct features for tag refinement, including user profile data, query data on Baidu Maps and basic properties of POIs. Finally, we conduct extensive experiments to demonstrate the effectiveness of our framework. And we further present a case study of the deployment of our framework on Baidu Maps.

Ranking query independent scholarly articles is a practical and difficult task, due to the heterogeneous, evolving and dynamic nature of entities involved in scholarly articles. To do this, we first propose a scholarly article ranking model by assembling the importance of involved entities (i.e., articles, venues and authors) such that the importance is a combination of prestige and popularity to capture the evolving nature of entities. To compute the prestige of articles and venues, we propose a novel Time-Weighted PageRank that extends traditional PageRank with a time decaying factor. We then develop a batch algorithm for scholarly article ranking, in which we propose a block-wise method for Time-Weighted PageRank in terms of an analysis of the citation characteristics of scholarly articles. We further develop an incremental algorithm for dynamic scholarly article ranking, which partitions graphs into affected and unaffected areas, and employs different updating strategies for nodes in different areas. Using real-life data, we finally conduct an extensive experimental study, and show that our approach is both effective and efficient for ranking scholarly articles.

Dense subgraph discovery has proven useful in various applications of temporal networks. We focus on a special class of temporal networks whose nodes and edges are kept fixed, but edge weights constantly and regularly vary with timestamps. However, finding dense subgraphs in temporal networks is nontrivial, and its state of the art solution uses a filter-and-verification framework, which is not scalable on large temporal networks. In this study, we propose a data-driven approach to finding dense subgraphs in large temporal networks with T timestamps. (1) We first develop a data-driven approach employing hidden statistics to identifying k time intervals, instead of T*(T + 1)/2 ones (k is typically much smaller than T), which strikes a balance between quality and efficiency. (2) After proving that the problem has no constant factor approximation algorithms, we design better heuristic algorithms to attack the problem, by building the connection of finding dense subgraphs with a variant of the Prize Collecting Steiner Tree problem. (3) Finally, we have conducted an extensive experimental study to demonstrate the effectiveness and efficiency of our approach, using real-life and synthetic data.

Network anomaly detection has become very popular in recent years because of the importance of discovering key regions of structural inconsistency in the network. In addition to application-specific information carried by anomalies, the presence of such structural inconsistency is often an impediment to the effective application of data mining algorithms such as community detection and classification. In this paper, we study the problem of detecting structurally inconsistent nodes that connect to a number of diverse influential communities in large social networks. We show that the use of a network embedding approach, together with a novel dimension reduction technique, is an effective tool to discover such structural inconsistencies. We also experimentally show that the detection of such anomalous nodes has significant applications: one is the specific use of detected anomalies, and the other is the improvement of the effectiveness of community detection.

A network with $n$ nodes contains O(n2) possible links. Even for networks of modest size, it is often difficult to evaluate all pairwise possibilities for links in a meaningful way. Furthermore, even though link prediction is closely related to missing value estimation problems, such as collaborative filtering, it is often difficult to use sophisticated models such as latent factor methods because of their computational complexity over very large networks. Due to this computational complexity, most known link prediction methods are designed for evaluating the link propensity over a specified subset of links, rather than for performing a global search over the entire networks. In practice, however, it is essential to perform an exhaustive search over the entire networks. In this paper, we propose an ensemble enabled approach to scaling up link prediction, which is able to decompose traditional link prediction problems into subproblems of smaller size. These subproblems are each solved with the use of latent factor models, which can be effectively implemented over networks of modest size. Furthermore, the ensemble enabled approach has several advantages in terms of performance. We show the advantage of using ensemble-based latent factor models with experiments on very large networks. Experimental results demonstrate the effectiveness and scalability of our approach.