Seiji Maekawa

Seiji Maekawa, Koki Noda, Yuya Sasaki et al.

Graph Neural Networks (GNNs) have achieved great success on a node classification task. Despite the broad interest in developing and evaluating GNNs, they have been assessed with limited benchmark datasets. As a result, the existing evaluation of GNNs lacks fine-grained analysis from various characteristics of graphs. Motivated by this, we conduct extensive experiments with a synthetic graph generator that can generate graphs having controlled characteristics for fine-grained analysis. Our empirical studies clarify the strengths and weaknesses of GNNs from four major characteristics of real-world graphs with class labels of nodes, i.e., 1) class size distributions (balanced vs. imbalanced), 2) edge connection proportions between classes (homophilic vs. heterophilic), 3) attribute values (biased vs. random), and 4) graph sizes (small vs. large). In addition, to foster future research on GNNs, we publicly release our codebase that allows users to evaluate various GNNs with various graphs. We hope this work offers interesting insights for future research.

Seiji Maekawa, Yuya Sasaki, George Fletcher et al.

We propose a framework that automatically transforms non-scalable GNNs into precomputation-based GNNs which are efficient and scalable for large-scale graphs. The advantages of our framework are two-fold; 1) it transforms various non-scalable GNNs to scale well to large-scale graphs by separating local feature aggregation from weight learning in their graph convolution, 2) it efficiently executes precomputation on GPU for large-scale graphs by decomposing their edges into small disjoint and balanced sets. Through extensive experiments with large-scale graphs, we demonstrate that the transformed GNNs run faster in training time than existing GNNs while achieving competitive accuracy to the state-of-the-art GNNs. Consequently, our transformation framework provides simple and efficient baselines for future research on scalable GNNs.

Seiji Maekawa, Yuya Sasaki, Makoto Onizuka

Node classification is one of the hottest tasks in graph analysis. Though existing studies have explored various node representations in directed and undirected graphs, they have overlooked the distinctions of their capabilities to capture the information of graphs. To tackle the limitation, we investigate various combinations of node representations (aggregated features vs. adjacency lists) and edge direction awareness within an input graph (directed vs. undirected). We address the first empirical study to benchmark the performance of various GNNs that use either combination of node representations and edge direction awareness. Our experiments demonstrate that no single combination stably achieves state-of-the-art results across datasets, which indicates that we need to select appropriate combinations depending on the dataset characteristics. In response, we propose a simple yet holistic classification method A2DUG which leverages all combinations of node representations in directed and undirected graphs. We demonstrate that A2DUG stably performs well on various datasets and improves the accuracy up to 11.29 compared with the state-of-the-art methods. To spur the development of new methods, we publicly release our complete codebase under the MIT license.

Catarina G. Belem, Pouya Pezeshkpour, Hayate Iso et al. · nvidia

Although many studies have investigated and reduced hallucinations in large language models (LLMs) for single-document tasks, research on hallucination in multi-document summarization (MDS) tasks remains largely unexplored. Specifically, it is unclear how the challenges arising from handling multiple documents (e.g., repetition and diversity of information) affect models outputs. In this work, we investigate how hallucinations manifest in LLMs when summarizing topic-specific information from multiple documents. Since no benchmarks exist for investigating hallucinations in MDS, we use existing news and conversation datasets, annotated with topic-specific insights, to create two novel multi-document benchmarks. When evaluating 5 LLMs on our benchmarks, we observe that on average, up to 75% of the content in LLM-generated summary is hallucinated, with hallucinations more likely to occur towards the end of the summaries. Moreover, when summarizing non-existent topic-related information, gpt-3.5-turbo and GPT-4o still generate summaries about 79.35% and 44% of the time, raising concerns about their tendency to fabricate content. To understand the characteristics of these hallucinations, we manually evaluate 700+ insights and find that most errors stem from either failing to follow instructions or producing overly generic insights. Motivated by these observations, we investigate the efficacy of simple post-hoc baselines in mitigating hallucinations but find them only moderately effective. Our results underscore the need for more effective approaches to systematically mitigate hallucinations in MDS. We release our dataset and code at github.com/megagonlabs/Hallucination_MDS.

Seiji Maekawa, Jackson Hassell, Pouya Pezeshkpour et al.

As language models gain access to external tools via structured function calls, they become increasingly more capable of solving complex, multi-step tasks. However, existing benchmarks for tool-augmented language models (TaLMs) provide insufficient control over factors such as the number of functions accessible, task complexity, and input size, and remain vulnerable to data contamination. We present FuncBenchGen, a unified, contamination-free framework that evaluates TaLMs by generating synthetic multi-step tool-use tasks. The key idea is to cast tool use as traversal over a hidden function-dependency DAG where nodes are function calls and an edge between nodes represents one function consuming the output of another. Given a set of external function schemas, initial variable values, and a target variable, models must compose the correct call sequence to compute the target variable. FuncBenchGen allows users to precisely control task difficulty (e.g., graph size, dependency depth, and distractor functions) while avoiding data leakage. We apply our FuncBenchGen framework to evaluate seven LLMs on tool use tasks of varying difficulty. Reasoning-optimized models consistently outperform general-purpose models with GPT-5 significantly outperforming other models. Performance declines sharply as dependency depth increases. Furthermore, connected irrelevant functions prove especially difficult to handle. We find that strong models often make syntactically valid function calls but propagate incorrect or stale argument values across steps, revealing brittle state tracking by LLMs in multi-turn tool use. Motivated by this observation, we introduce a simple mitigation strategy that explicitly restates prior variable values to the agent at each step. Surprisingly, this lightweight change yields substantial gains across models. e.g., yielding a success rate improvement from 62.5% to 81.3% for GPT-5.

Seiji Maekawa, Koh Takeuch, Makoto Onizuka

We consider the clustering problem of attributed graphs. Our challenge is how we can design an effective and efficient clustering method that precisely captures the hidden relationship between the topology and the attributes in real-world graphs. We propose Non-linear Attributed Graph Clustering by Symmetric Non-negative Matrix Factorization with Positive Unlabeled Learning. The features of our method are three holds. 1) it learns a non-linear projection function between the different cluster assignments of the topology and the attributes of graphs so as to capture the complicated relationship between the topology and the attributes in real-world graphs, 2) it leverages the positive unlabeled learning to take the effect of partially observed positive edges into the cluster assignment, and 3) it achieves efficient computational complexity, $O((n^2+mn)kt)$, where $n$ is the vertex size, $m$ is the attribute size, $k$ is the number of clusters, and $t$ is the number of iterations for learning the cluster assignment. We conducted experiments extensively for various clustering methods with various real datasets to validate that our method outperforms the former clustering methods regarding the clustering quality.