Large Language Models (LLMs) have demonstrated remarkable generalization capabilities, but aligning their outputs with human preferences typically requires expensive supervised fine-tuning. Recent test-time methods leverage textual feedback to overcome this, but they often critique and revise a single candidate response, lacking a principled mechanism to systematically analyze, weigh, and synthesize the strengths of multiple promising candidates. Such a mechanism is crucial because different responses may excel in distinct aspects (e.g., clarity, factual accuracy, or tone), and combining their best elements may produce a far superior outcome. This paper proposes the Textual Self-Attention Network (TSAN), a new paradigm for test-time preference optimization that requires no parameter updates. TSAN emulates self-attention entirely in natural language to overcome this gap: it analyzes multiple candidates by formatting them into textual keys and values, weighs their relevance using an LLM-based attention module, and synthesizes their strengths into a new, preference-aligned response under the guidance of the learned textual attention. This entire process operates in a textual gradient space, enabling iterative and interpretable optimization. Empirical evaluations demonstrate that with just three test-time iterations on a base SFT model, TSAN outperforms supervised models like Llama-3.1-70B-Instruct and surpasses the current state-of-the-art test-time alignment method by effectively leveraging multiple candidate solutions.

In real-world applications, spectral Graph Neural Networks (GNNs) are powerful tools for processing diverse types of graphs. However, a single GNN often struggles to handle different graph types—such as homogeneous and heterogeneous graphs—simultaneously. This challenge has led to the manual design of GNNs tailored to specific graph types, but these approaches are limited by the high cost of labor and the constraints of expert knowledge, which cannot keep up with the rapid growth of graph data. To overcome these challenges, we introduce AutoSGNN, an automated framework for discovering propagation mechanisms in spectral GNNs. AutoSGNN unifies the search space for spectral GNNs by integrating large language models with evolutionary strategies to automatically generate architectures that adapt to various graph types. Extensive experiments on nine widely-used datasets, encompassing both homophilic and heterophilic graphs, demonstrate that AutoSGNN outperforms state-of-the-art spectral GNNs and graph neural architecture search methods in both performance and efficiency.

In real-world scenarios, most complex systems can be generally modelled as homogenous or heterogenous networks. Therefore, downstream tasks (e.g., node/graph classification, node clustering) based on these two types of graphs become ubiquitous and have drawn considerable attentions in recent years. Existing literatures on node classification mainly focuses on either homogeneous or heterogeneous graphs, while research on effectively carrying out node classification tasks on both types of graphs simultaneously still under-exploited. To fill this gap, we propose a universal Graph Neural Network architecture based on Subgraph and Subhypergraph (SS-GNN) with feature-enhanced strategy for node embedding on both homogeneous and heterogeneous graphs. Through construction of subgraph and subhypergraph with same-class nodes, our model can simultaneously deal with homogeneous and heterogeneous graphs. Graph attention modules are especially designed to embed subgraphs of same-class nodes to learn the internal topological structure and local community structure within the original graph. Additionally, to capture high-order features of graph and enhance the embedding representations of nodes, we also utilize hypergraph attention modules to embed subhypergraphs of same-class nodes. Unlike other approaches that rely on pre-defined meta-paths, our model can be readily applied to most real-world applications without requiring any domain knowledge. Finally, we conduct extensive experiments on three homogeneous and three heterogeneous real-world graphs to demonstrate the effectiveness of SS-GNN. The experimental results for node classification and clustering tasks not only show the superior performance of our proposed model compared to state-of-the-art, but also demonstrate its potentially good interpretability for graph analysis. This work may provide some enlightening insights to the study on universality of graph foundation model.