Research

Research interests

I research representation learning on large-scale graphs via deep learning and applied machine learning techniques. Specifically, my research topics are:

Scalable representation learning under various graph models such as the knowledge graph, heterogeneous graphs, temporal graphs, static homogeneous graphs, etc.
Graph summarization to reduce the complexity of the large-scale real-world graph data while retaining the computational power of models in machine learning tasks.
Domain-specific graph analysis and discovery.

My research applications include:

Web data entity linkage in knowledge integration and knowledge base construction, i.e., consolidate records from different web sources that represent the same real-world entity, such as the same book listed in different online sales websites.
Temporal link prediction and anomaly detection in heterogeneous web networks, e.g., predicting the potential friend of a user in social networks even though they didn’t know about each other in the past, as well as detecting the fraud (such as a bot for malicious advertising) in user-product networks of the online sales website. This application aims to improve the quality of advertisement and customer purchasing experience.
Multi-modality representation fusion for recommendation & customization, e.g., how to incorporate the multi-modal information of a book (such as the image of its cover and the textual description) to better represent it for classification or advertising to customers?

Di Jin

Research interests

Scalable representation learning under various graph models such as the knowledge graph, heterogeneous graphs, temporal graphs, static homogeneous graphs, etc.

Graph summarization to reduce the complexity of the large-scale real-world graph data while retaining the computational power of models in machine learning tasks.

Domain-specific graph analysis and discovery.

Web data entity linkage in knowledge integration and knowledge base construction, i.e., consolidate records from different web sources that represent the same real-world entity, such as the same book listed in different online sales websites.

Multi-modality representation fusion for recommendation & customization, e.g., how to incorporate the multi-modal information of a book (such as the image of its cover and the textual description) to better represent it for classification or advertising to customers?