Pengfei Zhang

Mr. Pengfei Zhang


Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
10 January 2025 in Applied Sciences
Verified
Journal Article
Open access
PDF file
JPJiao Peng
Jiao Peng
YHYue He
Yue He
YCYongjuan Chang
Yongjuan Chang
YLYanyan Lu
Yanyan Lu
PZPengfei Zhang
Pengfei Zhang
QYQingzhi Yu
Qingzhi Yu
Multimodal sentiment analysis faces a number of challenges, including modality missing, modality heterogeneity gap, incomplete datasets, etc. Previous studies usually adopt schemes like meta-learning or multi-layer structures. Nevertheless, these methods lack interpretability for the interaction between modalities. In this paper, we constructed a new dataset, SM-MSD, for sentiment analysis in social media (SAS) that differs significantly from conventional corpora, comprising 10K instances of diverse data from Twitter, encompassing text, emoticons, emojis, and text embedded in images. This dataset aims to reflect authentic social scenarios and various emotional expressions, and provides a meaningful and challenging evaluation benchmark for multimodal sentiment analysis in specific contexts. Furthermore, we propose a multi-task framework based on heterogeneous graph neural networks (H-GNNs) and contrastive learning. For the first time, heterogeneous graph neural networks are applied to multimodal sentiment analysis tasks. In the case of additional labeling data, it guides the emotion prediction of the missing mode. We conduct extensive experiments on multiple datasets to verify the effectiveness of the proposed scheme. Experimental results demonstrate that our proposed scheme surpasses state-of-the-art methods by 1.7% and 0 in accuracy and 1.54% and 4.9% in F1-score on the MOSI and MOSEI datasets, respectively, and exhibits robustness to modality missing scenarios.

30 March 2025 in Applied Sciences
Verified
Journal Article
Open access
PDF file
JZJianbin Zhao
Jianbin Zhao
PZPengfei Zhang
Pengfei Zhang
YWYuzhen Wang
Yuzhen Wang
RXRui Xin
Rui Xin
XLXiuyuan Lu
Xiuyuan Lu
MSMeina Song
Meina Song
Table-based question answering (TableQA) has emerged as an important task in natural language processing, yet existing models face challenges in handling complex reasoning and mitigating hallucinations, especially when dealing with diverse table structures. We introduce TQAgent, a framework designed to enhance table-based reasoning by incorporating knowledge graphs and tree-structured reasoning paths. TQAgent reduces hallucinations and improves model reliability by grounding reasoning in external knowledge and dynamically sampling high-confidence paths. Additionally, it employs knowledge distillation techniques for lightweight deployment. Experimental results on the TabFact, WikiTQ, and FeTaQA datasets show significant performance improvements, with accuracy increases of up to 4% over baseline models. TQAgent’s dynamic operation planning and knowledge graph integration enable effective multi-step reasoning and better handling of diverse table data. Furthermore, the framework achieves state-of-the-art results, surpassing traditional large-scale models in both reasoning accuracy and computational efficiency. These findings open new avenues for future research in table-based question answering and model deployment optimization.

9 February 2026 in Algorithms
Verified
Journal Article
Open access
PDF file
JZJianbin Zhao
Jianbin Zhao
XXXing Xu
Xing Xu
SWShaoying Wang
Shaoying Wang
PZPengfei Zhang
Pengfei Zhang
SSShengyi Shen
Shengyi Shen
Yifan ZhuYifan Zhu
Yifan Zhu
Rotated object detection aims to achieve precise localization by strictly aligning bounding boxes with object orientations, thereby minimizing background interference. Existing methods predominantly focus on extracting intra-object features within rotated bounding boxes. However, these approaches often overlook the discriminative contextual information from the surrounding background, leading to classification ambiguity when internal features are indistinguishable. To address this limitation, we propose Background Information Fusion R-CNN (BIF-RCNN), a novel rotated object detection framework that strategically re-integrates the background context from the object’s horizontal enclosing region to validate its category, turning previously discarded “noise” into auxiliary discriminative cues. Specifically, we introduce a dual-level rotation-horizontal feature fusion module (DFM), which leverages horizontal bounding boxes enclosing the rotated objects to extract contextual background features. These features are then adaptively fused with the internal object features to enhance the overall representation capability of the model. In addition, we design a Prediction Difference and Entropy-Constrained Loss (PDE Loss), which guides the model to focus on hard-to-classify samples that are prone to confusion due to similar feature representations. This loss function improves the model’s robustness and discriminative power. Extensive experiments conducted on the DOTA benchmark dataset demonstrate the effectiveness of the proposed method. Notably, our approach achieves up to a 4.02% AP improvement in single-category detection performance compared to a strong baseline, highlighting its superiority in rotated object detection tasks.

12 March 2026 in Electronics
Verified
Journal Article
Open access
PDF file
XXXing Xu
Xing Xu
JZJianbin Zhao
Jianbin Zhao
PZPengfei Zhang
Pengfei Zhang
YLYaduo Liu
Yaduo Liu
BYBingyang Yu
Bingyang Yu
Yifan ZhuYifan Zhu
Yifan Zhu
Document-level Event Extraction (DEE) requires identifying complex event records and arguments dispersed across unstructured texts. However, applying general Large Language Models (LLMs) to DEE is intrinsically hindered by their lack of inductive bias for rigid structural constraints, often leading to schema violations and suboptimal performance in complex structural prediction tasks. To address this, we propose the S tructure-Aware Lightweight DEE, termed SALE, which leverages the structural reasoning potential of Code-Based LLMs (Code-LLMs) as a favorable inductive preference. We leverage the natural isomorphism between event schemas and programming object definitions, formulating event extraction as a Python 3.9 class instantiation task to bridge the gap between semantic understanding and structural adherence. Specifically, SALE employs a novel two-stage training paradigm: First, a Structure-Aware Fine-tuning stage injects general structural knowledge via diverse code-style instruction tasks derived from broad Information Extraction (IE) datasets; second, an Event Extraction Alignment stage utilizes a reward-based alignment loss—optimized via policy gradient—to adapt this capability to document-level intricacies. The effectiveness of SALE stems from the synergy between its structure-aware prompting and the specialized alignment stage built on a code-oriented backbone. Extensive experiments on established news-domain benchmarks (RAMS and WikiEvents) demonstrate that our approach significantly outperforms representative supervised and general LLM baselines in cross-task zero-shot and few-shot transfer settings (e.g., surpassing supervised baselines by over 7% in F1 score). Furthermore, SALE maintains a highly efficient inference profile and parameter-efficient footprint, offering a practical and scalable solution for vertical domain applications.