CV

Zhijin Guo

Email: Zhijin.Guo@bristol.ac.uk
Website: https://zhijinguo.github.io/

Research Interest

• Natural Language Processing
• Knowledge Graph Embedding
• Graph Neural Network
• Social Network Analysis
• Recommender System

Education

• Ph.D., University of Bristol (2021 – Present)
  • Supervisor: Nello Cristianini, Martha Lewis, Edwin Simpson
  • Thesis: Embedding Relational Data: Analysis, Methods, Applications
• M.Sc., Advanced Computer Science, The University of Sheffield (2019 – 2020)
  • GPA: Distinction
• B.Sc., Software Engineering, Northeastern University (2015 – 2019)

Employment History

• 2023 – Present: Research Assistant, Twitter Data Analysis, University of Bristol
• 2021 – 2023: Teaching Assistant, Introduction to Data Analytics (NLP), University of Bristol

Research Publications

1. Z. Guo, Z. Xu, M. Lewis, and N. Cristianini, “Compositional Fusion of Signals in Data Embedding”, arXiv preprint arXiv:2311.11085, 2023.
2. Z. Guo, Z. Xu, M. Lewis, and N. Cristianini, “Extract: Explainable Transparent Control of Bias in Embeddings”, 2023.
3. Z. Xu, Z. Guo, and N. Cristianini, “On compositionality in data embedding,” pp. 484–496, 2023.

Funding

• 2023 Jean Golding Institute Seed Corn Funding

Research Projects

• 2023-2024 Social Network Analysis
  • Investigating the impact of medical influencers on social media health discourse, addressing misinformation and public trust challenges.
  • Utilizing prompt learning for LLMs to build a multi-label classifier of influencers’ identities and their network actors based on social media profiles and content.
  • Utilizing LLMs for topic modelling in influencers’ posts, providing insights into their narrative strategies.
  • Exploring relationships between influencers’ identities, their social network connections, and discursive frames using Graph Neural Networks (GNNs) for in-depth socio-semantic network analysis.
• 2023 Understanding Compositionality in Data Embedding
  • Developed two novel methods for analyzing three types of data embeddings: Correlation-based Compositionality Detection and Additive Compositionality Detection.
  • Demonstrated that word embeddings encapsulate both semantic and morphological information, and sentence embeddings can be interpreted as aggregations of subject, verb, and objects.
  • Applied techniques to knowledge graph embeddings, revealing the ability to infer node attributes not included in training, such as demographics in user embeddings of a recommender system.
• 2022-2023 Harmonic Graphs: Leveraging Graph Neural Networks for Predictive Modeling of Relational Music N-grams
  • Applied Graph Neural Networks to analyze symbolic elements in music scores, specifically in polyphonic music.
  • Used graph embeddings to explore musical similarities and relationships, categorizing compositions by style, genre, and borrowing practices.
  • Analyzed the contextual use and connections of musical themes (soggetti) to understand compositional patterns and influences.
• 2022 EXTRACT: Explainable Transparent Control of Bias in Embeddings
  • Developed EXTRACT, a suite of methods for controlling bias in knowledge graph embeddings, enhancing transparency and explainability.
  • Demonstrated through experiments on the MovieLens-1M dataset that personal attributes like gender, age, and occupation can be inferred from user viewing behaviors. Conducted further experiments on the KG20C citation dataset, revealing the ability to deduce publication conference information from article citation networks.
  • Introduced four transparent methods to balance embedding effectiveness for intended predictions while minimizing retention of unwanted information, highlighting the trade-off between accuracy and privacy.