Knowledge Graph Construction by LLM and for LLM

PhD Qualifying Examination


Title: "Knowledge Graph Construction by LLM and for LLM"

by

Mr. Jiaxin BAI


Abstract:

Language models have seen remarkable advancements in natural language 
processing, but effective utilization often lacks structured knowledge 
representation. Knowledge graphs, which capture relationships between entities 
and, more recently, events and activities, offer immense potential for 
improving language understanding and reasoning capabilities. This survey 
focuses on two key aspects: (1) the automatic construction of knowledge graphs 
leveraging large language models and (2) enhancing language models by 
integrating structured knowledge from knowledge graphs. The first part explores 
techniques for automatically constructing knowledge graphs using large language 
models. It first investigates how large language models can be incorporated 
into traditional knowledge graph construction pipelines, such as entity 
recognition, relation extraction, and event extraction. Then, it investigates 
how LLMs can be used to construct other semantic graphs parallel to the 
traditional knowledge graph construction pipelines. The second part delves into 
strategies for utilizing structured knowledge graphs to enhance language 
models' performance. It examines how incorporating knowledge graphs can improve 
capabilities in tasks like information retrieval, question answering, knowledge 
reasoning, etc. The survey discusses techniques for integrating knowledge 
graphs into language models and highlights the benefits of structured knowledge 
representation.

This survey presents an overview of existing research, methodologies, and 
evaluation metrics related to knowledge graph construction and integration with 
language models. It examines recent advancements, identifies open challenges, 
and discusses potential future directions.


Date:                   Tuesday, 20 August 2024

Time:                   1:00pm - 3:00pm

Venue:                  Room 5501
                        Lifts 25/26

Committee Members:      Dr. Yangqiu Song (Supervisor)
                        Dr. Xiaojuan Ma (Chairperson)
                        Dr. Long Chen
                        Dr. Junxian He
                        Dr. Dongdong She