In recent years, we mainly focus on transfer learning in different scenarios, and we have led 2 PhDs and 10 masters in this direction to actively carry out related research work in cross-domain classification, cross-language word vector representation and cross-language entity alignment.

I. Transfer Learning:

Transfer learning aims to use the source domain knowledge to guide the solution of target domain problems, which has become an important research direction in the field of artificial intelligence and data mining due to its ability to effectively overcome the problem of obtaining a large number of labels and model re-training in traditional machine learning, and is considered to be one of the next hot technologies that may be successfully applied in business after deep learning.

Our team has been working on the research of transfer learning in different data scenarios for many years, which mainly involves cross-domain classification, cross-language vector representation learning and so on. He has published more than 30 papers as the first author or corresponding author, participated in writing a monograph on data stream classification, and granted 2 software copyrights and 5 patents.

2. Graph-based machine learning:

Network is prevalent in practical applications, such as social networks, information networks, biological networks, and knowledge graphs. Social shopping networks contain many types of objects such as users, goods, shops, etc. The relationship between objects is no longer only purchase, but contains finer interactions such as collection and favourite. Based on these finer information, more accurate knowledge discovery results can be generated. Obviously, compared with the existing independent distribution data that have been widely studied and homogeneous networks (i.e., nodes or edges in the network have the same type), heterogeneous networks provide more in-depth modelling of both the connotative representations of nodes and the relational representations between nodes, and contain richer structural and semantic information, which in turn provides a new, more accurate and interpretable way for knowledge discovery. How to mine this information is of great value for the development of our economy, medicine, education and other fields.

3. Entity Alignment of Knowledge Graph

Knowledge graph is a giant semantic web composed of nodes and edges, in which the nodes represent entities and the edges represent various semantic relationships between entities, which is a new, more effective and comprehensive representation, and has received widespread attention for its large-scale, interpretable, and reasonable characteristics, and has been applied in a number of fields, such as search engines [3], automated Q&A [4], and interpretable recommendations [5]. However, knowledge graph construction requires a large corpus, insufficient automation of the construction process, and different data sources lead to data omissions, errors, obsolescence, and conflicts in knowledge graphs, etc. Therefore, providing a unified view of data through knowledge graph fusion in order to expand the scale of the graphs and improve the quality of the graphs has become an important research topic.

No content

No content

No content