A5024599321- Topic Modeling
- Natural Language Processing Techniques
- Advanced Text Analysis Techniques
- Speech and dialogue systems
- Multimodal Machine Learning Applications
- Sentiment Analysis and Opinion Mining
- Biomedical Text Mining and Ontologies
- Language, Metaphor, and Cognition
- Speech Recognition and Synthesis
- Music and Audio Processing
- Domain Adaptation and Few-Shot Learning
- Text and Document Classification Technologies
- Semantic Web and Ontologies
- Data Quality and Management
- Advanced Graph Neural Networks
- Music Technology and Sound Studies
- Computational and Text Analysis Methods
- Complex Network Analysis Techniques
- Text Readability and Simplification
- Mental Health via Writing
- Software Engineering Research
- Generative Adversarial Networks and Image Synthesis
- Spam and Phishing Detection
- Recommender Systems and Techniques
- Artificial Intelligence in Games
University of Manchester
2023-2025
Guangzhou Medical University
2025
Third Affiliated Hospital of Guangzhou Medical University
2025
University of Sheffield
2018-2024
Zhejiang University
2019-2024
Fujian Electric Power Survey & Design Institute
2024
University of Surrey
2023
Ghent University Hospital
2023
Durham University
2023
Shenzhen University
2021
Sentiment analysis or opinion mining aims to use automated tools detect subjective information such as opinions, attitudes, and feelings expressed in text. This paper proposes a novel probabilistic modeling framework based on Latent Dirichlet Allocation (LDA), called joint sentiment/topic model (JST), which detects sentiment topic simultaneously from Unlike other machine learning approaches classification often require labeled corpora for classifier training, the proposed JST is fully...
Sentiment analysis or opinion mining aims to use automated tools detect subjective information such as opinions, attitudes, and feelings expressed in text. This paper proposes a novel probabilistic modeling framework called joint sentiment-topic (JST) model based on latent Dirichlet allocation (LDA), which detects sentiment topic simultaneously from A reparameterized version of the JST Reverse-JST, obtained by reversing sequence generation process, is also studied. Although equivalent...
In this paper, we investigate the social-based recommendation algorithms on heterogeneous social networks and proposed Hete-CF, a collaborative filtering algorithm using relations. Distinct from exiting methods, Hete-CF can effectively utilise multiple types of relations in network. More importantly, is general approach be used arbitrary networks, including event based location any other information associated with information. The experimental results real-world dataset DBLP (a typical...
Metaphoric expressions are widespread in natural language, posing a significant challenge for various language processing tasks such as Machine Translation. Current word embedding based metaphor identification models cannot identify the exact metaphorical words within sentence. In this paper, we propose an unsupervised learning method that identifies and interprets metaphors at word-level without any preprocessing, outperforming strong baselines task. Our model extends to interpret...
End-to-end training with Deep Neural Networks (DNN) is a currently popular method for metaphor identification. However, standard sequence tagging models do not explicitly take advantage of linguistic theories We experiment two DNN which are inspired by human identification procedures. By testing on three public datasets, we find that our achieve state-of-the-art performance in end-to-end
Lay summarisation aims to jointly summarise and simplify a given text, thus making its content more comprehensible non-experts.Automatic approaches for lay can provide significant value in broadening access scientific literature, enabling greater degree of both interdisciplinary knowledge sharing public understanding when it comes research findings. However, current corpora this task are limited their size scope, hindering the development broadly applicable data-driven approaches. Aiming...
Interactive Natural Language Processing (iNLP) has emerged as a novel paradigm within the field of NLP, aimed at addressing limitations in existing frameworks while aligning with ultimate goals artificial intelligence. This considers language models agents capable observing, acting, and receiving feedback iteratively from external entities. Specifically, this context can: (1) interact humans for better understanding user needs, personalizing responses, human values, improving overall...
Self-supervised learning (SSL) has recently emerged as a promising paradigm for training generalisable models on large-scale data in the fields of vision, text, and speech. Although SSL been proven effective speech audio, its application to music audio yet be thoroughly explored. This is partially due distinctive challenges associated with modelling musical knowledge, particularly tonal pitched characteristics music. To address this research gap, we propose an acoustic Music undERstanding...
Social media data are produced continuously by a large and uncontrolled number of users. The dynamic nature such requires the sentiment topic analysis model to be also dynamically updated, capturing most recent language use sentiments topics in text. We propose Joint Sentiment-Topic (dJST) which allows detection tracking views current recurrent interests shifts sentiment. Both dynamics captured assuming that sentiment-topic-specific word distributions generated according at previous epochs....
Recognising dialogue acts (DA) is important for many natural language processing tasks such as generation and intention recognition. In this paper, we propose a dual-attention hierarchical recurrent neural network DA classification. Our model partially inspired by the observation that conversational utterances are normally associated with both topic, where former captures social act latter describes subject matter. However, dependency between DAs topics has not been utilised most existing...
Visually analyzing citation networks poses challenges to many fields of the data mining research. How can we summarize a large graph according user's interest? In particular, how illustrate impact highly influential paper through summarization? Can maintain sensory node-link structure while revealing flow-based influence patterns and preserving fine readability? The state-of-the-art maximization algorithms detect most node in network, but fail account for its influence. On other hand,...
In this paper, we propose FrameBERT, a BERT-based model that can explicitly learn and incorporate FrameNet Embeddings for concept-level metaphor detection. FrameBERT not only achieves better or comparable performance to the state-of-the-art, but also is more explainable interpretable compared existing models, attributing its ability of accounting external knowledge FrameNet.
As the demand for highly secure and dependable lightweight systems increases in modern world, Physically Unclonable Functions (PUFs) continue to promise a alternative high-cost encryption techniques key storage. While security features promised by PUFs are attractive system designers, they have been shown be vulnerable various sophisticated attacks - most notably Machine Learning (ML) based modelling (ML-MA) which attempt digitally clone PUF behaviour thus undermine their security. More...
Biomedical factoid question answering is an important task in biomedical applications. It has attracted much attention because of its reliability. In systems, better representation words great importance, and proper word embedding can significantly improve the performance system. With success pretrained models general natural language processing tasks, have been widely used areas, many model-based approaches proven effective question-answering tasks. addition to embedding, name entities also...
Medical dialogue generation aims to generate responses according a history of turns between doctors and patients. Unlike open-domain generation, this requires background knowledge specific the medical domain. Existing generative frameworks for fall short incorporating domain-specific knowledge, especially with regard terminology. In paper, we propose novel framework improve by considering features centered on We leverage an attention mechanism incorporate terminologically centred features,...
Large language models (LLMs) achieved remarkable performance across various tasks. However, they face challenges in managing long documents and extended conversations, due to significantly increased computational requirements, both memory inference time, potential context truncation when the input exceeds LLM’s fixed length. This paper proposes a method called Selective Context that enhances efficiency of LLMs by identifying pruning redundancy make more compact. We test our approach using...
Music composition represents the creative side of humanity, and itself is a complex task that requires abilities to understand generate information with long dependency harmony constraints. While demonstrating impressive capabilities in STEM subjects, current LLMs easily fail this task, generating ill-written music even when equipped modern techniques like In-Context-Learning Chain-of-Thoughts. To further explore enhance LLMs' potential by leveraging their reasoning ability large knowledge...
Yi Cheng, Siyao Li, Bang Liu, Ruihui Zhao, Sujian Chenghua Lin, Yefeng Zheng. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.
Incorporating external graph knowledge into neural chatbot models has been proven effective for enhancing dialogue generation. However, in conventional networks (GNNs), message passing on a is independent from text, resulting the representation hidden space differing that of text. This training regime existing therefore leads to semantic gap between and In this study, we propose novel framework enhanced We dynamically construct multi-hop with pseudo nodes involve language model feature...