PhD Studentship in Computer Science: Mitigating Hallucinations of Large Language Models (LLMs) with Knowledge Graphs based Recommendation Techniques

Award Summary

100% fees covered, and a minimum tax-free annual living allowance of £18,622 (2023/24 UKRI rate). SAgE commits additional in‐kind contribution with software resources, data, training, supervision and travel support.

Overview

Hallucination is a common problem of large language models (LLMs). Knowledge graph based recommendation techniques can effectively represent, organize, utilize, and validate both general and domain knowledge. This project will explore how to mitigate hallucination problem of LLMs with knowledge graph based recommendation techniques. The methodology includes a theoretical and an experimental component. The theoretical component explores the mitigation approaches from the aspects of 1) novel knowledge injection, alignment, and maintenance approaches; 2) novel model architecture, in-context learning, neural symbolic reasoning, retrieval augmented generation techniques; 3) novel evaluation framework to validate responses from different sources including LLMs, human feedback, and recommendation engine. The experimental component is to conduct evaluation experiments of the proposed approaches on large scale data from novels, movies, and online communities as well as the impact case studies in finance domain provided by industry partner SAGE.

Number Of Awards

1

Start Date

September 2024

Award Duration

4 years

Application Closing Date

29.02.2024

Sponsor

SAGE (www.sage.com )

Supervisors

Dr Huizhi Liang

Dr Deepayan Bhowmik

Dr Srijith Rajamohan

Eligibility Criteria

You must have, or expect to gain, a minimum 2:1 Honours degree or international equivalent in a subject relevant to the proposed PhD project (inc. computing, mathematics, engineering etc.). Enthusiasm for research, the ability to think and work independently, excellent analytical skills and strong verbal and written communication skills are also essential requirements.

Home and international applicants (inc. EU) are welcome to apply and if successful will receive a full studentship. Applicants whose first language is not English require an IELTS score of 6.5 overall with a minimum of 5.5 in all sub-skills.

International applicants may require an ATAS (Academic Technology Approval Scheme ) clearance certificate prior to obtaining their visa and to study on this programme. 

How To Apply

You must apply through the University’s Apply to Newcastle Portal  

Once registered select ‘Create a Postgraduate Application’.  

Use ‘Course Search’ to identify your programme of study:  

• search for the ‘Course Title’ using the programme code: 8050F
• Research Area: Computing Science
• select ‘PhD Computer Science (full time) - Computing Science' as the programme of study 

You will then need to provide the following information in the ‘Further Details’ section:  

• ‘Personal Statement’ (this is a mandatory field) - upload a document or write a statement directly in to the application form
• When prompted regarding your research proposal - select ‘Write Proposal’. You should then type in the title of the research project from this advert. You do not need to upload a research proposal in this field.  
• The studentship code COMP2157 in the ‘Studentship/Partnership Reference’ field  
• Degree transcripts and certificates and, if English is not your first language, a copy of your English language qualification if already completed.

In the ‘Supporting Documentation’ section please upload:

Contact Details

Dr Huizhi Liang