Dr Delvin Ce Zhang recently joined the School of Computer Science, so we asked some questions to get to know him a little better and understand more about his work, past experiences and what he will be bringing to our School.
What are your key areas of research?
DZ: My research broadly lies in Natural Language Processing and Machine Learning. Specifically, I focus on applying Multi-Modal Large Language Models for many real-world problems, including Misinformation Detection, Question Answering, Recommender Systems, and AI for Science.
These days, I am particularly interested in Misinformation Detection. The advancements of generative language models enable them to produce answers and relevant information to users’ questions. However, the information from large language models may not always be accurate due to their outdated or incomplete knowledge. As a result, false information or misinformation can mislead the public, causing unnecessary worries. This underscores the urgent need for developing robust misinformation detection or fact-checking methods to automatically verify the correctness of the generated information, ultimately providing accurate and reliable information to users.
Did something in particular draw you to this research field?
Why did you want to pursue a career in academia and, in particular, in Computer Science?
DZ: My motivation to pursue a career in academia begins with my mother, who works as a high school teacher in China. I am always influenced by her dedication to teaching and learning over these years, thus I choose to work in academia to inherit her occupation and her enthusiasm for education. Within academia, I have opportunities to continue learning, teaching, and researching.
After studying for a PhD degree in Computer Science, I am further inspired by my PhD supervisor, who keeps exploring new concepts, solving challenging problems, and making research contributions. Computer Science is a subject that allows me to use mathematics and programming to solve real-world problems, which eventually benefits the overall society in a meaningful way. I am particularly interested in what model architecture to design and how to use real-world data to train the model to uncover the latent patterns behind the observed data. Such a pattern discovery process is usually beyond human cognitive capability, and I am always curious about what insightful discovery I get from such a process. This is why I continue my career in Computer Science.
Do you have any recent publications that you would like to highlight?
DZ: One of my recent publications, which appeared on NAACL-25, studies a critical problem, fact-checking or misinformation detection. Nowadays, fact-checking methods rely on reliable sources, such as academic papers, to verify the correctness of the given information. In our NAACL-25 work, we find that simply relying on the textual content within an academic paper as a reliable source is insufficient, since it might contain coreferential expressions and acronyms, which are unclear and difficult for a computer to understand. To solve this problem, we discovered that incorporating cited papers as additional context helps, since cited papers usually discuss similar content and can resolve coreferential expressions and acronyms in most cases. By analyzing both the textual content within the paper and its cited papers, we propose a graph-based language model, which reaches a higher fact-checking accuracy than existing models.
I highlight this work for two main reasons. First, fact-checking or misinformation detection is a crucial problem that every online user should be aware of, so that they are not misled by wrong information. Second, this work opens a new avenue for future research, i.e., incorporating multi-modal data, such as images and knowledge graphs, for a more comprehensive fact-checking.
What is your favourite thing about teaching the next generation of computer scientists?
In terms of practice-based learning, Computer Science requires students to put theoretical knowledge into practice, so that students can gain an insightful understanding through hands-on exercise. I will encourage students to conduct lab tests in the class and do a course project after the class. For lab tests, I aim to ask students to first present their solutions, then I provide feedback immediately to correct their answers for an effective learning. For the course project, I plan to select some outstanding projects and ask students to share and discuss their solutions with the rest, with the goal of creating a peer-teaching and knowledge-sharing environment, with appropriate guidance and advice from the instructor.
What attracted you to working at The University of Sheffield?
DZ: The University of Sheffield is famous for its research and teaching, and its NLP group is especially reputable in the research community. The dynamic and active environment of the group attracts me a lot, and I expect to have research discussions and collaborations with colleagues here.
And finally, do you have any hobbies or interests that you would like to share?
DZ: Outside of work, I quite enjoy running, hiking, and watching table tennis. Peak District National Park is a famous place, and I look forward to exploring it and taking photos of its wonderful views. The European Table Tennis Championships are regularly held within the continent, and I expect to watch the live games soon.