Min Peng

Knowledge Graph Question Answering (KGQA) aims to answer natural language questions by reasoning over structured knowledge graphs (KGs). While large language models (LLMs) have advanced KGQA through their strong reasoning capabilities, existing methods continue to struggle to fully exploit both the rich knowledge encoded in KGs and the reasoning capabilities of LLMs, particularly in complex scenarios. They often assume complete KG coverage and lack mechanisms to judge when external information is needed, and their reasoning remains locally myopic, failing to maintain coherent multi-step planning, leading to reasoning failures even when relevant knowledge exists. We propose Graph-RFT, a novel two-stage reinforcement fine-tuning KGQA framework with a ''plan–KGsearch–and–Websearch–during–think'' paradigm, that enables LLMs to perform autonomous planning and adaptive retrieval scheduling across KG and web sources under incomplete knowledge conditions. Graph-RFT introduces a chain-of-thought (CoT) fine-tuning method with a customized plan–retrieval dataset activates structured reasoning and resolves the GRPO cold-start problem. It then introduces a novel plan–retrieval guided reinforcement learning process integrates explicit planning and retrieval actions with a multi-reward design, enabling coverage-aware retrieval scheduling. It employs a Cartesian-inspired planning module to decompose complex questions into ordered sub-questions, and logical expression to guide tool invocation for globally consistent multi-step reasoning. This reasoning–retrieval process is optimized with a multi-reward combining outcome and retrieval-specific signals, enabling the model to learn when and how to combine KG and web retrieval effectively. Experiments on multiple KGQA benchmarks demonstrate that Graph-RFT achieves superior performance over strong baselines, even with smaller LLM backbones, and substantially improves complex question decomposition, factual coverage, and tool coordination.

With the rapid growth of visual content, automated aesthetic evaluation has become increasingly important. However, existing research faces three key challenges: (1) the absence of datasets combining Image Aesthetic Assessment (IAA) scores and Image Aesthetic Captioning (IAC) descriptions; (2) limited integration of quantitative scores and qualitative text, hindering comprehensive modeling; (3) the subjective nature of aesthetics, which complicates consistent fine-grained evaluation. To tackle these issues, we propose a unified multimodal framework. To address the lack of data, we develop the Textual Aesthetic Sentiment Labeling Pipeline (TASLP) for automatic annotation and construct the Reddit Multimodal Sentiment Dataset (RMSD) with paired IAA and IAC labels. To improve annotation integration, we introduce the Aesthetic Category Sentiment Analysis (ACSA) task, which models fine-grained aesthetic attributes across modalities. To handle subjectivity, we design two models—LAGA for IAA and ACSFM for IAC—that leverage ACSA features to enhance consistency and interpretability. Experiments on RMSD and public benchmarks show that our approach alleviates data limitations and delivers competitive performance, highlighting the effectiveness of fine-grained sentiment modeling and multimodal learning in aesthetic evaluation.

AI has found extensive application in edge analytics, with Deep Neural Networks effectively processing and analyzing data. Some applications need to deal with unstructured data like graphs. This complexity and diversity of unstructured features pose challenges for traditional DNN models. Therefore, Graph Neural Networks have emerged as a significant solution. However, GNNs encounter two primary challenges in edge analytics scenarios: scalability and efficiency. These challenges arise primarily from significant structural differences among various graphs and models. To address these issues, we propose AGF, an Adaptive GNN Framework. AGF consists of three components: subgraph partitioning and model partitioning, scalable hardware architecture, and design space exploration. Subgraph partitioning and model partitioning transform complex and diverse GNNs into a uniform computing flow and subgraphs. These can be directly mapped onto the hardware kernel, leveraging GNN characteristics to achieve independent parallelism. The hardware architecture is based on a multi-level processing element structure, which can efficiently parallelize subgraphs and vertices based on the partitioning. The design space exploration selects kernel allocation strategies and setups based on the hardware, dataset, and model. We validated AGF using six datasets and three GNN models. The results demonstrate that our framework achieves an acceleration ranging from 4.54 to 53.17× compared to the GPU-based GNN framework. When compared to other FPGA-based GNN accelerators, AGF achieves a latency reduction ranging from 1.16 to 33.6× in model execution and from 1.09 to 2.37× in end-to-end scenarios.