Hongzhi Qi

Dr. Hongzhi Qi

orcid

Research Focus & Metrics

Academic Career Length

8Years (2019-Present)

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

Representative Publications

View All Publications
19 August 2025 in Bioengineering
Verified
Journal Article
Open access
PDF file
HQHongzhi Qi
Hongzhi Qi
GFGuanghui FU
Guanghui FU
JLJianqiang Li
Jianqiang Li
CSChangwei Song
Changwei Song
WZWei Zhai
Wei Zhai
QZQing Zhao
Qing Zhao
On social media, users often express their personal feelings, which may exhibit cognitive distortions or even suicidal tendencies on certain specific topics. Early recognition of these signs is critical for effective psychological intervention. In this paper, we introduce two novel datasets from Chinese social media: SOS-HL-1K for suicidal risk classification, which contains 1249 posts, and SocialCD-3K, a multi-label classification dataset for cognitive distortion detection that contains 3407 posts. We conduct a comprehensive evaluation using two supervised learning methods and eight large language models (LLMs) on the proposed datasets. From the prompt engineering perspective, we experiment with two types of prompt strategies, including four zero-shot and five few-shot strategies. We also evaluate the performance of the LLMs after fine-tuning on the proposed tasks. Experimental results show a significant performance gap between prompted LLMs and supervised learning. Our best supervised model achieves strong results, with an F1-score of 82.76% for the high-risk class in the suicide task and a micro-averaged F1-score of 76.10% for the cognitive distortion task. Without fine-tuning, the best-performing LLM lags by 6.95 percentage points in the suicide task and a more pronounced 31.53 points in the cognitive distortion task. Fine-tuning substantially narrows this performance gap to 4.31% and 3.14% for the respective tasks. While this research highlights the potential of LLMs in psychological contexts, it also shows that supervised learning remains necessary for more challenging tasks.

24 November 2023
Verified
Conference Paper
Closed access
PHPing He
Ping He
HQHongzhi Qi
Hongzhi Qi
JCJiayue Cang
Jiayue Cang
The objective of an aspect-based sentiment analysis task is to identify the emotional polarity of specific aspects in text sequences. Previous study usually only involve the text modality. However, user-generated comment data generally include not only plain text, but also a large number of image and text data. For ignoring the cross-modal correlation between image and multiple aspects of text, and the problem of alignment between text and image high-level semantics. We propose a new aspect-based cross-modal sentiment analysis model HCMI-PLNT (Hierarchical Cross-Modal Interaction - Pre-LN Transformer). The core idea of this model is to use the hierarchical cross-modal Pre-LN Transformer interaction module to capture the bidirectional deep interaction between aspect, text and image. The high-level semantic feature of images are obtained by extracting adjective-noun pairs (ANPs) from images, and the text is aligned with the high-level semantic of images. The accuracy of HCMI-PLNT model on the TWITTER-15 data set is 78.60%, and the accuracy on the TWITTER-17 data set is 72.43%. The experiment proves the feasibility and effectiveness of the HCMI-PLNT model.

1 January 2019
Verified
Dataset
Closed access
QLQi Lu
Qi Lu
CDCuiyu Deng
Cuiyu Deng
LFLi Fu
Li Fu
RWRui Wu
Rui Wu
LCLianxia Chang
Lianxia Chang
YZYue Zhao
Yue Zhao

14 October 2024 in arXiv
Verified
Preprint
Open access
PDF file
WZWei Zhai
Wei Zhai
NBNan Bai
Nan Bai
QZQing Zhao
Qing Zhao
JLJianqiang Li
Jianqiang Li
FWFan Wang
Fan Wang
GFGuanghui FU
Guanghui FU
As the prevalence of mental health challenges, social media has emerged as a key platform for individuals to express their emotions.Deep learning tends to be a promising solution for analyzing mental health on social media. However, black box models are often inflexible when switching between tasks, and their results typically lack explanations. With the rise of large language models (LLMs), their flexibility has introduced new approaches to the field. Also due to the generative nature, they can be prompted to explain decision-making processes. However, their performance on complex psychological analysis still lags behind deep learning. In this paper, we introduce the first multi-task Chinese Social Media Interpretable Mental Health Instructions (C-IMHI) dataset, consisting of 9K samples, which has been quality-controlled and manually validated. We also propose MentalGLM series models, the first open-source LLMs designed for explainable mental health analysis targeting Chinese social media, trained on a corpus of 50K instructions. The proposed models were evaluated on three downstream tasks and achieved better or comparable performance compared to deep learning models, generalized LLMs, and task fine-tuned LLMs. We validated a portion of the generated decision explanations with experts, showing promising results. We also evaluated the proposed models on a clinical dataset, where they outperformed other LLMs, indicating their potential applicability in the clinical field. Our models show strong performance, validated across tasks and perspectives. The decision explanations enhance usability and facilitate better understanding and practical application of the models. Both the constructed dataset and the models are publicly available via: https://github.com/zwzzzQAQ/MentalGLM.

29 June 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
QHQinbo Hui
Qinbo Hui
JLJianqiang Li
Jianqiang Li
JLJinjia Li
Jinjia Li
CSChangwei Song
Changwei Song
MZMuhammad Zubair
Muhammad Zubair
HQHongzhi Qi
Hongzhi Qi
Task scheduling in large-scale parallel simulations, wherein each task is formally represented as a simulation model, is inherently constrained by complex Directed Acyclic Graph (DAG) dependencies, which dictate rigid execution sequences and induce severe synchronization blocking. Traditional meta-heuristic scheduling paradigms often suffer from prohibitive optimization overheads and rigidly force models to passively await predecessor availability, resulting in severe resource starvation and elongated per-step makespans. To resolve this bottleneck, this paper proposes an intelligent expert system centered on a novel Zero-Step Decoupling Strategy, comprising two core components: First, a knowledge-driven Zero-Step Decoupling Strategy, which leverages K-Means clustering for optimal resource matching ($\mathcal{O}(|\mathcal{M}|)$ complexity) and transforms rigid spatial DAG dependencies into asynchronous inter-step temporal fetches. This mechanism allows all models to achieve absolute pipeline concurrency from the first physical simulation step, effectively replacing structural blocking with pipeline execution. Second, an Adaptive Model Migration mechanism realizes dynamic fault-tolerance and intelligent load re-equilibration. By leveraging the decoupling of the Zero-Step strategy, migrated models can "hot-swap" across heterogeneous virtual machines (VMs) with near-zero topological re-synchronization overhead, ensuring system-wide robustness. Extensive empirical evaluations on a heterogeneous distributed testbed featuring 1,000 concurrent models demonstrate that the proposed expert system significantly outperforms traditional paradigms. Specifically, it compresses scheduling overhead to 29.3 ms, maintains an outstanding CPU saturation of 88.5%, and restricts the passive wait time ratio to 1.1%, thereby delivering a speedup ratio of 2.86 and reducing the overall makespan by 60.7% compared to the gold-standard HEFT algorithm. This intelligent, robust, and deterministic expert system provides a critical foundation for advanced, high-availability industrial simulation applications.

4 June 2026 in IEEE Transactions on Knowledge and Data Engineering
Verified
Journal Article
Closed access
HQHongzhi Qi
Hongzhi Qi
JLJianqiang Li
Jianqiang Li
YGYanhu Ge
Yanhu Ge
HWHuimin Wu
Huimin Wu
YCYuqi Cai
Yuqi Cai
CCChaojin Chen
Chaojin Chen
Graph-based Retrieval-Augmented Generation (GraphRAG) is a prominent technique for mitigating hallucinations in Large Language Models (LLMs), yet it faces critical limitations in high-stakes anesthesiology question answering. At the retrieval stage, pervasive synonymy and knowledge fragmentation cause clinical queries to resolve to incorrect synonym clusters, while the highly specialized and domain-specific nature of clinical language widens the semantic gap, making an effective balance between recall and precision difficult to achieve. At the reasoning stage, the high structural heterogeneity of anesthesiology knowledge graphs, combined with the strong context-dependence of clinical semantics, renders existing multi-hop reasoning susceptible to clinical logic drift along weak associations, thereby introducing unreliable reasoning paths. Finally, at the fusion stage, conflicts between retrieved evidence and the model's parametric knowledge can easily lead to highly confident yet clinically unsafe hallucinations. To address these challenges, we propose the Description-augmented, Uncertainty-aware, and Adaptive Layered Knowledge Framework (DUAL-Know). For retrieval, query augmentation and description-augmented semantic recall bridge terminology gaps and comprehensively retrieve relevant knowledge. For reasoning, a Dynamically Gated Heterogeneous Multi-Head Attention (DGHMA) mechanism accommodates graph heterogeneity and dynamically injects query intent into each reasoning layer, followed by an uncertainty-aware path ranking module that filters noisy chains. For fusion, a multi-metric scoring strategy jointly evaluates model confidence, retrieval consistency, and semantic coherence to safely arbitrate knowledge conflicts. Experiments demonstrate that DUAL-Know consistently outperforms strong baselines, generating more accurate, reliable, and verifiable answers for complex clinical question-answering tasks.