A5101990229- Topic Modeling
- Natural Language Processing Techniques
- Sentiment Analysis and Opinion Mining
- Advanced Text Analysis Techniques
- Text and Document Classification Technologies
- Text Readability and Simplification
- Speech Recognition and Synthesis
- Multimodal Machine Learning Applications
- Speech and dialogue systems
- Biomedical Text Mining and Ontologies
- Handwritten Text Recognition Techniques
- Semantic Web and Ontologies
- Domain Adaptation and Few-Shot Learning
- Second Language Acquisition and Learning
- Anomaly Detection Techniques and Applications
- Data Quality and Management
- Mental Health via Writing
- COVID-19 diagnosis using AI
- Chemotherapy-related skin toxicity
- Geophysical Methods and Applications
- Graphite, nuclear technology, radiation studies
- Advanced Graph Neural Networks
- Public Relations and Crisis Communication
- Rough Sets and Fuzzy Logic
- Parallel Computing and Optimization Techniques
Chinese Academy of Medical Sciences & Peking Union Medical College
2025
Sanya Central Hospital
2022-2025
Soochow University
2015-2024
Institute of Art
2024
Yangtze River Pharmaceutical Group (China)
2024
Hunan University of Science and Technology
2023
Tencent (China)
2023
Heilongjiang University
2009-2019
Heilongjiang University of Science and Technology
2007-2018
Northeast Agricultural University
2015
Extracting biomedical entities and their relations from text has important applications on research. Previous work primarily utilized feature-based pipeline models to process this task. Many efforts need be made feature engineering when are employed. Moreover, may suffer error propagation not able utilize the interactions between subtasks. Therefore, we propose a neural joint model extract as well simultaneously, it can alleviate problems above.Our was evaluated two tasks, i.e., task of...
Neural networks have shown promising results for relation extraction. State-of-the-art models cast the task as an end-to-end problem, solved incrementally using a local classifier. Yet previous work statistical demonstrated that global optimization can achieve better performances compared to classification. We build globally optimized neural model extraction, proposing novel LSTM features in order learn context representations. In addition, we present method integrate syntactic information...
Meishan Zhang, Zhenghua Li, Guohong Fu, Min Zhang. Proceedings of the 2019 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). 2019.
Joint models have shown stronger capabilities for Chinese word segmentation and POS tagging, received great interests in the community of natural language processing. In this paper, we follow line work, presenting a simple yet effective sequence-to-sequence neural model joint task, based on well-defined transition system, by using long short term memory network structures. We conduct experiments five different datasets. The results demonstrate that our proposed is highly competitive. By...
Emotion recognition in multi-party conversation (ERMC) is becoming increasingly popular as an emerging research topic natural language processing. Prior focuses on exploring sequential information but ignores the discourse structures of conversations. In this paper, we investigate importance handling informative contextual cues and speaker-specific features for ERMC. To end, propose a discourse-aware graph neural network (ERMC-DisGCN) particular, design relational convolution to lever...
Meishan Zhang, Yue Guohong Fu. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and 9th International Joint (EMNLP-IJCNLP). 2019.
Semantic role labeling (SRL), also known as shallow semantic parsing, is an important yet challenging task in NLP. Motivated by the close correlation between syntactic and structures, traditional discrete-feature-based SRL approaches make heavy use of features. In contrast, deep-neural-network-based usually encode input sentence a word sequence without considering structures. this work, we investigate several previous for encoding trees, thorough study on whether extra syntax-aware...
Representation learning is the foundation of machine reading comprehension and inference. In state-of-the-art models, character-level representations have been broadly adopted to alleviate problem effectively representing rare or complex words. However, character itself not a natural minimal linguistic unit for representation word embedding composing due ignoring coherence consecutive characters inside word. This paper presents general subword-augmented framework computationally derived...
This paper presents a lexicalized HMM-based approach to Chinese named entity recognition (NER). To tackle the problem of unknown words, we unify word identification and NER as single tagging task on sequence known words. do this, first employ known-word bigram-based model segment sentence into then apply uniformly HMMs assign each proper hybrid tag that indicates its pattern in forming an category formed entity. Our system is able integrate both internal formation patterns surrounding...
While part-of-speech (POS) tagging and dependency parsing are observed to be closely related, existing work on joint modeling with manually crafted feature templates suffers from the sparsity incompleteness problems. In this paper, we propose an approach POS using transition-based neural networks. Three network based classifiers designed resolve shift/reduce, tagging, labeling conflicts. Experiments show that our significantly outperforms previous methods for across a variety of natural languages.
To the Editor: Chemotherapy is one of mainstay therapies for patients with breast cancer (BC),[1,2] but it often restricted by dose-limiting toxicities attributed to myelosuppression. Although recombinant human granulocyte colony-stimulating factors (rhG-CSF) are effective treating and preventing chemotherapy-induced neutropenia, daily administration inconvenient patients.[3,4] Telpegfilgrastim (Peijin®, Xiamen Amoytop Biotech Co., Ltd, Xiamen, China) a Y-shape branched pegylated factor...
Recent studies in Large Vision-Language Models (LVLMs) have demonstrated impressive advancements multimodal Out-of-Context (OOC) misinformation detection, discerning whether an authentic image is wrongly used a claim. Despite their success, the textual evidence of images retrieved from inverse search directly transmitted to LVLMs, leading inaccurate or false information decision-making phase. To this end, we present E2LVLM, novel evidence-enhanced large vision-language model by adapting two...
Nuclear instance segmentation and classification provide critical quantitative foundations for digital pathology diagnosis. With the advent of foundational Segment Anything Model (SAM), accuracy efficiency nuclear have improved significantly. However, SAM imposes a strong reliance on precise prompts, its class-agnostic design renders results entirely dependent provided prompts. Therefore, we focus generating prompts with more accurate localization propose \textbf{APSeg},...
Pre-trained language models (PLMs) have shown great potentials in natural processing (NLP) including rhetorical structure theory (RST) discourse parsing.Current PLMs are obtained by sentence-level pre-training, which is different from the basic unit, i.e. element unit (EDU).To this end, we propose a second-stage EDU-level pre-training approach work, presents two novel tasks to learn effective EDU representations continually based on well pre-trained models.Concretely, (1) next prediction...
In this paper, we take morphemes as the basic tokens and present a fine-to-coarse strategy for Chinese sentence-level sentiment classification. This study involves three parts. First, employ morphological productivity to extract from dictionary calculate their polarity intensity at same time. Then, apply acquired morpheme-level information predict semantic orientation of words phrases within an opinionated sentence. Finally, all scores are combined determine The experimental results on...
Relation classification is associated with many potential applications in the artificial intelligence area. Recent approaches usually leverage neural networks based on structure features such as syntactic or dependency to solve this problem. However, high-cost make inconvenient be directly used. In addition, are probably domain-dependent. Therefore, paper proposes a bi-directional long-short-term-memory recurrent-neural-network (Bi-LSTM-RNN) model low-cost sequence address relation...