Understanding and reasoning over complex spreadsheets remain fundamental challenges for large language models (LLMs), which often struggle with intricate structures and rely solely on neural computation. In this work, we propose SheetBrain, a neuro-symbolic dual-workflow agent framework for precise and interpretable reasoning over tabular data. SheetBrain consists of an understanding module that produces a comprehensive overview of the spreadsheet, including structural summaries and query-specific analyses to guide execution; an execution module that integrates a Python sandbox with preloaded table-processing libraries and an Excel helper toolkit for effective data manipulation; and a validation module that verifies the correctness of reasoning and answers, triggering re-execution if necessary. We evaluate SheetBrain on multiple public QA and manipulation benchmarks, and introduce SheetBench, a new benchmark targeting large, multi-table, and structurally complex spreadsheets. Experimental results show that SheetBrain significantly improves reasoning performance on both existing benchmarks and the more challenging scenarios presented in SheetBench.

Hanbing Liu, Lang Cao, Yuanyi Ren, Mengyu Zhou, Haoyu Dong, Xiaojun Ma, Shi Han, Dongmei Zhang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Spreadsheets are critical to data-centric tasks, with rich, structured layouts that enable efficient information transmission.Given the time and expertise required for manual spreadsheet layout design, there is an urgent need for automated solutions.However, existing automated layout models are ill-suited to spreadsheets, as they often (1) treat components as axis-aligned rectangles with continuous coordinates, overlooking the inherently discrete, gridbased structure of spreadsheets; and (2) neglect interrelated semantics, such as data dependencies and contextual links, unique to spreadsheets.In this paper, we first formalize the spreadsheet layout generation task, supported by a seven-criterion evaluation protocol and a dataset of 3,326 spreadsheets.We then introduce SheetDesigner, a zero-shot and trainingfree framework using Multimodal Large Language Models (MLLMs) that combines rule and vision reflection for component placement and content population.SheetDesigner outperforms five baselines by at least 22.6%.We further find that through vision modality, MLLMs handle overlap and balance well but struggle with alignment, necessitates hybrid rule and visual reflection strategies.Our codes and data is available at Github.

Large Language Models (LLMs) demonstrate exceptional performance in textual understanding and tabular reasoning tasks. However, their ability to comprehend and analyze hybrid text, containing textual and tabular data, remains unexplored. The hybrid text often appears in the form of hybrid long documents (HLDs), which far exceed the token limit of LLMs. Consequently, we apply an Automated Information Extraction framework (AIE) to enable LLMs to process the HLDs and carry out experiments to analyse four important aspects of information extraction from HLDs. Given the findings: 1) The effective way to select and summarize the useful part of a HLD. 2) An easy table serialization way is enough for LLMs to understand tables. 3) The naive AIE has adaptability in many complex scenarios. 4) The useful prompt engineering to enhance LLMs on HLDs. To address the issue of dataset scarcity in HLDs and support future work, we also propose the Financial Reports Numerical Extraction (FINE) dataset. The dataset and code are publicly available in the attachments.

Tabular data analysis is crucial in various fields, and large language models show promise in this area. However, current research mostly focuses on rudimentary tasks like Text2SQL and TableQA, neglecting advanced analysis like forecasting and chart generation. To address this gap, we developed the Text2Analysis benchmark, incorporating advanced analysis tasks that go beyond the SQL-compatible operations and require more in-depth analysis. We also develop five innovative and effective annotation methods, harnessing the capabilities of large language models to enhance data quality and quantity. Additionally, we include unclear queries that resemble real-world user questions to test how well models can understand and tackle such challenges. Finally, we collect 2249 query-result pairs with 347 tables. We evaluate five state-of-the-art models using three different metrics and the results show that our benchmark presents introduces considerable challenge in the field of tabular data analysis, paving the way for more advanced research opportunities.

Graph Transformer is gaining increasing attention in the field of machine learning and has demonstrated state-of-the-art performance on benchmarks for graph representation learning. However, as current implementations of Graph Transformer primarily focus on learning representations of small-scale graphs, the quadratic complexity of the global self-attention mechanism presents a challenge for full-batch training when applied to larger graphs. Additionally, conventional sampling-based methods fail to capture necessary high-level contextual information, resulting in a significant loss of performance. In this paper, we introduce the Hierarchical Scalable Graph Transformer (HSGT) as a solution to these challenges. HSGT successfully scales the Transformer architecture to node representation learning tasks on large-scale graphs, while maintaining high performance. By utilizing graph hierarchies constructed through coarsening techniques, HSGT efficiently updates and stores multi-scale information in node embeddings at different levels. Together with sampling-based training methods, HSGT effectively captures and aggregates multi-level information on the hierarchical graph using only Transformer blocks. Empirical evaluations demonstrate that HSGT achieves state-of-the-art performance on large-scale benchmarks with graphs containing millions of nodes with high efficiency.

Collaborative filtering (CF) is an important research direction in recommender systems that aims to make recommendations given the information on user-item interactions. Graph CF has attracted more and more attention in recent years due to its effectiveness in leveraging high-order information in the user-item bipartite graph for better recommendations. Specifically, recent studies show the success of graph neural networks (GNN) for CF is attributed to its low-pass filtering effects. However, current researches lack a study of how different signal components contributes to recommendations, and how to design strategies to properly use them well. To this end, from the view of spectral transformation, we analyze the important factors that a graph filter should consider to achieve better performance. Based on the discoveries, we design JGCF, an efficient and effective method for CF based on Jacobi polynomial bases and frequency decomposition strategies. Extensive experiments on four widely used public datasets show the effectiveness and efficiency of the proposed methods, which brings at most 27.06% performance gain on Alibaba-iFashion. Besides, the experimental results also show that JGCF is better at handling sparse datasets, which shows potential in making recommendations for cold-start users.

Tables are widely used for data storage and presentation due to their high flexibility in layout. The importance of tables as information carriers and the complexity of tabular data understanding attract a great deal of research on large-scale pre-training for tabular data. However, most of the works design models for specific types of tables, such as relational tables and tables with well-structured headers, neglecting tables with complex layouts. In real-world scenarios, there are many such tables beyond their target scope that cannot be well supported. In this paper, we propose GetPt, a unified pre-training architecture for general table representation applicable even to tables with complex structures and layouts. First, we convert a table to a heterogeneous graph with multiple types of edges to represent the layout of the table. Based on the graph, a specially designed transformer is applied to jointly model the semantics and structure of the table. Second, we devise the Alternate Attention Network (AAN) to better model the contextual information across multiple granularities of a table including tokens, cells, and the table. To better support a wide range of downstream tasks, we further employ three pre-training objectives and pre-train the model on a large table dataset. We fine-tune and evaluate GetPt model on two representative tasks, table type classification, and table structure recognition. Experiments show that GetPt outperforms existing state-of-the-art methods on these tasks.

With the rapid development of the World Wide Web (WWW), heterogeneous graphs (HG) have explosive growth. Recently, heterogeneous graph neural network (HGNN) has shown great potential in learning on HG. Current studies of HGNN mainly focus on some HGs with strong homophily properties (nodes connected by meta-path tend to have the same labels), while few discussions are made in those that are less homophilous. Recently, there have been many works on homogeneous graphs with heterophily. However, due to heterogeneity, it is non-trivial to extend their approach to deal with HGs with heterophily. In this work, based on empirical observations, we propose a meta-path-induced metric to measure the homophily degree of a HG. We also find that current HGNNs may have degenerated performance when handling HGs with less homophilous properties. Thus it is essential to increase the generalization ability of HGNNs on non-homophilous HGs. To this end, we propose HDHGR, a homophily-oriented deep heterogeneous graph rewiring approach that modifies the HG structure to increase the performance of HGNN. We theoretically verify HDHGR. In addition, experiments on real-world HGs demonstrate the effectiveness of HDHGR, which brings at most more than 10% relative gain.

Graph Transformers have proved their advantages in graph data mining with elaborate Positional Encodings, especially in graph-level tasks. However, their application in the node classification task has not been fully exploited yet. In the node classification task, existing Graph Transformers with Positional Encodings are limited by the following issues: (i) PEs describing the node’s positional identities are insufficient for the node classification task on complex graphs, where a full portrayal of the local node property is needed. (ii) PEs for graphs are integrated with Transformers in a constant schema, resulting in the ignorance of local patterns that may vary among different nodes. In this paper, we propose Adaptive Graph Transformer (AGT) to tackle above issues. AGT consists of a Learnable Centrality Encoding and a Kernelized Local Structure Encoding. The two modules extract structural patterns from centrality and subgraph views in a learnable and scalable manner. Further, we design the Adaptive Transformer Block to adaptively integrate the attention scores and Structural Encodings in a node-specific manner. AGT achieves state-of-the-art performances on nine real-world web graphs (up to 1.6 million nodes). Furthermore, AGT shows outstanding results on two series of synthetic graphs with ranges of heterophily and noise ratios.

Graph Neural Networks (GNNs) are widely used on a variety of graph-based machine learning tasks. For node-level tasks, GNNs have strong power to model the homophily property of graphs (i.e., connected nodes are more similar), while their ability to capture heterophily property is often doubtful. This is partially caused by the design of the feature transformation with the same kernel for the nodes in the same hop and the followed aggregation operator. One kernel cannot model the similarity and the dissimilarity (i.e., the positive and negative correlation) between node features simultaneously even though we use attention mechanisms like Graph Attention Network (GAT), since the weight calculated by attention is always a positive value. In this paper, we propose a novel GNN model based on a bi-kernel feature transformation and a selection gate. Two kernels capture homophily and heterophily information respectively, and the gate is introduced to select which kernel we should use for the given node pairs. We conduct extensive experiments on various datasets with different homophily-heterophily properties. The experimental results show consistent and significant improvements against state-of-the-art GNN methods.

Graph Neural Networks (GNNs) show strong expressive power on graph data mining, by aggregating information from neighbors and using the integrated representation in the downstream tasks. The same aggregation methods and parameters for each node in a graph are used to enable the GNNs to utilize the homophily relational data. However, not all graphs are homophilic, even in the same graph, the distributions may vary significantly. Using the same convolution over all nodes may lead to the ignorance of various graph patterns. Furthermore, many existing GNNs integrate node features and structure identically, which ignores the distributions of nodes and further limits the expressive power of GNNs. To solve these problems, we propose Meta Weight Graph Neural Network (MWGNN) to adaptively construct graph convolution layers for different nodes. First, we model the Node Local Distribution (NLD) from node feature, topological structure and positional identity aspects with the Meta-Weight. Then, based on the Meta-Weight, we generate the adaptive graph convolutions to perform a node-specific weighted aggregation and boost the node representations. Finally, we design extensive experiments on real-world and synthetic benchmarks to evaluate the effectiveness of MWGNN. These experiments show the excellent expressive power of MWGNN in dealing with graph data with various distributions.

Researches on analyzing graphs with Graph Neural Networks (GNNs) have been receiving more and more attention because of the great expressive power of graphs. GNNs map the adjacency matrix and node features to node representations by message passing through edges on each convolution layer. However, the message passed through GNNs is not always beneficial for all parts in a graph. Specifically, as the data distribution is different over the graph, the receptive field (the farthest nodes that a node can obtain information from) needed to gather information is also different. Existing GNNs treat all parts of the graph uniformly, which makes it difficult to adaptively pass the most informative message for each unique part. To solve this problem, we propose two regularization terms that consider message passing locally: (1) Intra-Energy Reg and (2) Inter-Energy Reg. Through experiments and theoretical discussion, we first show that the speed of smoothing of different parts varies enormously and the topology of each part affects the way of smoothing. With Intra-Energy Reg, we strengthen the message passing within each part, which is beneficial for getting more useful information. With Inter-Energy Reg, we improve the ability of GNNs to distinguish different nodes. With the proposed two regularization terms, GNNs are able to filter the most useful information adaptively, learn more robustly and gain higher expressiveness. Moreover, the proposed LEReg can be easily applied to other GNN models with plug-and-play characteristics. Extensive experiments on several benchmarks verify that GNNs with LEReg outperform or match the state-of-the-art methods. The effectiveness and efficiency are also empirically visualized with elaborate experiments.

The abundant information in graphs helps us to learn more expressive node representations. Different nodes in the neighborhood have different importance to the central node. Thus, average weight aggregation in most Graph Neural Networks would fail to model such difference. GAT-based models introduce the attention mechanism to solve this problem, but they ignore the rich structural information and may suffer from the problem of over-smoothing. In this paper, we propose Graph Neural Networks with STructural Adaptive Receptive fields (STAR-GNN), which adaptively construct a receptive field for each node with structural information and further achieve better aggregation of information. Firstly, we model local structural distribution based on anonymous random walks, followed by using the structural information to construct receptive fields guided with mutual information. Then, as the generated receptive fields are irregular, we design a sub-graph aggregator to boost node representations and theoretically prove that it has the ability to capture the complex structures in receptive fields. Experimental results demonstrate the power of STAR-GNN in learning structural receptive fields adaptively and encoding more informative structural characteristics in real-world networks.

Information networks are ubiquitous and are ideal for modeling relational data. Networks being sparse and irregular, network embedding algorithms have caught the attention of many researchers, who came up with numerous embeddings algorithms in static networks. Yet in real life, networks constantly evolve over time. Hence, evolutionary patterns, namely how nodes develop itself over time, would serve as a powerful complement to static structures in embedding networks, on which relatively few works focus. In this paper, we propose EPNE, a temporal network embedding model preserving evolutionary patterns of the local structure of nodes. In particular, we analyze evolutionary patterns with and without periodicity and design strategies correspondingly to model such patterns in time-frequency domains based on causal convolutions. In addition, we propose a temporal objective function which is optimized simultaneously with proximity ones such that both temporal and structural information are preserved. With the adequate modeling of temporal information, our model is able to outperform other competitive methods in various prediction tasks. © 2020 The authors and IOS Press.

Keyword-based tags (referred to as tags) are used to represent additional attributes of nodes in addition to what is explicitly stated in their contents, like the hashtags in YouTube. Aside of being auxiliary information for node representation, tags can also be used for retrieval, recommendation, content organization, and event analysis. Therefore, tag representation learning is of great importance. However, to learn satisfactory tag representations is challenging because 1) traditional representation methods generally fail when it comes to representing tags, 2) bidirectional interactions between nodes and tags should be modeled, which are generally not dealt within existing research works. In this paper, we propose a tag representation learning model which takes tag-related node interaction into consideration, named Tag2Gauss. Specifically, since tags represent node communities with intricate overlapping relationships, we propose that Gaussian distributions would be appropriate in modeling tags. Considering the bidirectional interactions between nodes and tags, we propose a tag representation learning model mapping tags to distributions consisting of two embedding tasks, namely Tag-view embedding and Node-view embedding. Extensive evidence demonstrates the effectiveness of representing tag as a distribution, and the advantages of the proposed architecture in many applications, such as the node classification and the network visualization.

Chinese medical concept normalization, which maps non-standard medical concepts to standard expressions, is a NLP task with wide-ranging applications in medical big data research and clinical statistic. Many previous works apply supervised methods which require a lot of annotated data. However, they can not address the challenge brought by the high cost of medical data annotation, which requires sufficient professional knowledge and experience. Meanwhile, existing unsupervised methods perform poorly facing the various non-standard expression from different data sources. In this paper, we propose DUNE, Disease Unsupervised Normalization by Embedding, an unsupervised Chinese medical concept normalization framework by applying denoising auto-encoder (DAE) and network embedding. We formulate this task as finding mention-entity pairs with great text and comorbidity similarity. To handle the noise in text, we design a multi-view attention based denoising auto-encoder (MADAE) to capture text information from multiple views, reduce the influence of noise, and transform text to denoised vectors. To introduce comorbidity information, we construct a comorbidity network with both standard and non-standard disease names as nodes from medical records. Because of the diversity of nonstandard expressions, one disease perhaps corresponds to several different nodes, which causes noise in comorbidity network. To handle such network structure noise, we propose a denoising network embedding framework, which reduce the structure noise with the help of text information, to embed the nodes to vectors for comorbidity similarity measurement. Convincing experiment results show that our method performs better than existing unsupervised baselines and approaches the performance of classical supervised machine learning model on this task.