Zan Qiu

Dr. Zan Qiu

orcid

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

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

Representative Publications

View All Publications
5 November 2024 in Information
Verified
Journal Article
Open access
PDF file
ZQZan Qiu
Zan Qiu
GHGuimin Huang
Guimin Huang
XQXingguo Qin
Xingguo Qin
YWYabing Wang
Yabing Wang
JWJiahao Wang
Jiahao Wang
YZYa Zhou
Ya Zhou
The accuracy of traditional topic models may be compromised due to the sparsity of co-occurring vocabulary in the corpus, whereas conventional word embedding models tend to excessively prioritize contextual semantic information and inadequately capture domain-specific features in the text. This paper proposes a hybrid semantic representation method that combines a topic model that integrates conceptual knowledge with a weighted word embedding model. Specifically, we construct a topic model incorporating the Probase concept knowledge base to perform topic clustering and obtain topic semantic representation. Additionally, we design a weighted word embedding model to enhance the contextual semantic information representation of the text. The feature-based information fusion model is employed to integrate the two textual representations and generate a hybrid semantic representation. The hybrid semantic representation model proposed in this study was evaluated based on various English composition test sets. The findings demonstrate that the model presented in this paper exhibits superior accuracy and practical value compared to existing text representation methods.

31 December 2024 in Electronics
Verified
Journal Article
Open access
PDF file
YLYiqun Li
Yiqun Li
YZYa Zhou
Ya Zhou
ZQZan Qiu
Zan Qiu
YWYabing Wang
Yabing Wang
JWJiahao Wang
Jiahao Wang
GHGuimin Huang
Guimin Huang
Automatic speech recognition (ASR) technology has reached a mature level, and improving performance in data-scarce scenarios has become a key research focus. In this study, we propose a novel approach to building acoustic models by leveraging the multi-head attention mechanism to enhance various acoustic features extracted from audio data, combined with time-delay neural networks (TDNNs) using semi-orthogonal low-rank matrix factorization, resulting in significant performance improvements. To increase the robustness of the acoustic model, we introduce a small amount of data perturbation during the early stages of training. Additionally, during the decoding phase, we employ an external recurrent neural network language model (RNNLM) for rescoring, further enhancing the model’s accuracy. A comprehensive evaluation of the widely used LibriSpeech corpus shows that our method, using only half the training data, achieves word error rates (WERs) of 3.15 and 7.04 on the test_clean and test_other datasets, respectively. These results outperform models with similar architectures trained on the full dataset and demonstrate performance comparable to mainstream end-to-end models.

13 October 2026 in Electronics
Verified
Journal Article
Open access
PDF file
ZQZan Qiu
Zan Qiu
GHGuimin Huang
Guimin Huang
YLYiqun Li
Yiqun Li
YWYabing Wang
Yabing Wang
In short text classification tasks, existing studies primarily focus on extracting textual information while neglecting the semantics and dependencies among category labels. In practical applications, label information can effectively alleviate the semantic sparsity issue of short texts. This paper proposes a short text classification model that integrates label-feature augmentation with a multi-attention mechanism to deeply explore the dependencies between label semantics and text classification. First, the model expands labels via external knowledge bases to augment label representations. Meanwhile, for preprocessed short texts, entity linking is implemented to obtain corresponding entity information, which is then employed to retrieve relevant conceptual information from external knowledge bases, thereby uncovering deeper semantic content. Furthermore, a multi-attention mechanism is designed to learn the feature relationships among different embeddings generated by pre-trained models: a text self-attention module extracts key text representations, a label-text attention module captures text-related label features, an entity-text attention module retrieves associated entity information, and a concept-text attention module extracts conceptual-level semantics. Finally, these features are fused and input into a fully connected network to capture multi-level semantic information and complete short text classification. Experimental results on public benchmark datasets demonstrate that the proposed model significantly outperforms comparison methods on short text classification tasks and achieves superior performance.