About Me
I am currently a third-year PhD student at the University of Technology Sydney (UTS), advised by Dr. Shoujin Wang and co-supervised by Distinguished Prof. Fang Chen. Prior to that, I received my M.S. and B.Eng. degrees from the School of Computer Science and Technology at China University of Mining and Technology (CUMT), where I was advised by Prof. Guan Yuan. I also closely collaborate with Dr. Peilin Zhou from the New York University Abu Dhabi (NYUAD).
I have published 10+ papers at top international conferences and journals, including ICLR, SIGIR, TOIS, EMNLP, WWWJ, ICME, TCSS and PAKDD. My research focuses on building efficient and trustworthy AI systems under real-world scenarios. This question drives my work across three directions:
- Recommender Systems: My core research area, inlcuding sequential recommendaiton, generative recommendaiton and credible content recommendation (mitigating the recommendaiton of uncredible content, such as fake news, misinformation and other unreliable content). I am increasingly focused on how to build efficient generative recommendation to facilitate their real-world application, including sing-step GR.
- Generative Models: I study how generative models (diffusion models, flow matching) can be adapted in RSs, inlcuding sequential recommendation (using diffusion models to improve the diversity of sequential recommendation) and generative recommendation (uisng flow matching to improve the efficiency of generative recommendaiton).
- Large Language Models: Study the causal reasoning ability and trustworthiness of LLMs. I am currently exploring utilizing LLMs to improve the content credibility in RSs and provide reliable explanation.
I am actively seeking job opportunities in industry in the areas of RSs and GR. If you are interested in academic collaboration or would like to discuss potential opportunities, please feel free to email me at tobelbl2022@gmail.com.
🔥 News
📝 Publications
Survey Papers
We present a systematic survey of existing efforts to enhance the causal reasoning ability of large language models, covering domain-knowledge-driven methods (domain experts, contextual knowledge, predefined prompts, and fine-tuning) and model-driven methods (causal graph construction, causal effect estimation, and counterfactual reasoning).
We present a comprehensive survey of trustworthy large language model-based recommender systems, organized around the dimensions of fairness, robustness, privacy, explainability, and hallucination, with a detailed comparison of the strengths, weaknesses, and applicable scenarios of each category of methods.
Technical Papers
A single-step generative recommendation model that decode SID tokens in a single step. It view SID generation as a flow of token generation probabilities and characterize it by its average velocity over the whole generation process.
A generative recommender that improve recommendation crediblity while preserving high recommendation accuracy, consisting of a credibility-aware tokenizer and a accuracy-preserving and credibility-oriented discrete diffusion generator.
We propose MisEraser, an unlearning framework for removing uncredible content in recommendation. The framework consists of a user-preference–uncredible-content disentanglement network, an uncredible-feature fusion network, and an uncredible-content erasure network, which jointly achieve lossless erasure of uncredible content.
A diffusion-based content recommender that improves recommendation credibility while preserving high recommendation accuracy, consisting of a disentangled diffusion model that separates uncredible content from user preference signals and a progressively enhanced credible subspace projection that suppresses known and unknown uncredible content via null-space projection.
We propose DiffDiv, a diffusion-based sequential recommendation model for diversified recommendation. It constructs diversity-aware conditions to guide the diffusion training process and designs a heterogeneous diffusion inference to capture users' multi-faceted interests, further improving recommendation diversity.
📖 Education
- 2024.04-present, PhD. In Data Science Institute, UTS
- 2020.09-2023.06, M.S. In Computer Science and Technology, CUMT
- 2016.09-2020.06, B.Eng. In Computer Science and Technology, CUMT
🎖️ Honors and Awards
- Google Travel Grant
- Australian Government Scholarship (IRTP)
- National Scholarship by Ministry of Education of the People’s Republic of China
- Outstanding Graduate Student of CUMT
- Outstanding Graduation Thesis of CUMT