Heterogeneous, Sequel-aware, LLM-enhanced Graph Neural Networks

A research project is available in the Department of Management Science and Systems, which focuses on the development of heterogeneous, sequel-aware, LLM-enhanced GNNs.

Graph-based recommendation systems use higher-order user and item embeddings for next-item predictions. Dynamically adding collaborative signals from neighbors helps to use similar users' preferences during learning. While item-item correlations and their impact on recommendations have been studied, the efficacy of temporal item sequences for recommendations is much less explored. This research aims to explore problems in this domain. The work is algorithmic in nature and requires extensive benchmarking with state-of-the-art algorithms, tinkering with algorithm designs and related modifications, and implementation on a real-world case study.

The undergraduate student(s) who are expected to join the project will help process data collected from recommendation systems by careful extraction of features, performing statistical analysis, and finding how they affect the performance of the overall models for sequential prediction. In particular, they will be involved with the following activities:

a. Become familiar with the overall architecture of the project, including the multimodal data collected.
b. Have prior knowledge or learn about sequential models for recommendation, such as attention networks, and others
c. Be familiar with Natural Language Processing techniques (including Large Language Models, LLMs) and use tools that are developed for processing them and integrate them into our sequential learning algorithms.
d. Will be involved in reading research papers, interpreting them, and using implementations as available from GitHub repositories.
e. Enthusiastically participate in brown bag sessions and group presentations.
f. Expected to work with graduate students (Master's and PhD) and undergraduate students who are involved in the project. Teamwork is highly encouraged, although some tasks will have to be accomplished individually. There will be group meetings every week to discuss project-related and unrelated issues.

The ideal student would work with us for at least TWO semesters (preferably more!), including winter or summer break as appropriate.
What we offer: This is an UNFUNDED project. However, we do have the opportunity to do any of the following:

i) For UB students - Offer independent study/thesis credits as appropriate.
ii) For industry partners, this could be a way to participate in more rigorous, long-term, academic projects to keep abreast of state-of-the-art technology in the realm of recommendation systems
iii) Recommendation letters, ONLY after the student has made substantial contributions to the project, worked for at least two semesters, and ideally has co-authored a paper with any of the faculty members involved in the project.
iv) For external collaborators, there exist multiple opportunities for discussing how you can be involved with and contribute to our research (such as new projects and data sharing, problems that are of interest to industry, funding opportunities for new work, and more!).

The projects are available immediately and are very hands-on, fast-paced, with a lot of opportunity to develop skills pertaining to real-world data processing, programming (Python), visualization, use of machine learning knowledge acquired in the curriculum. It is anticipated that the project will continue for several years.

If the above is of interest, please send your resume and a few paragraphs on why this is of interest to you and what you hope to accomplish through your participation to Professor Dutta.

• Research papers or technical reports.
• Detailed codebase, which will be updated in GitHub repo.
• If accepted, presentations at conferences and workshops.

• Sequential learning algorithm design and enhancement.
• Empirical analysis and benchmarking with state-of-the-art algorithms.
• Work on a real-world case study using sequential learning.

• SEAS and CAS at UB
• India-based non-profit 

graph neural networks, recommendation systems, management science and systems, python, machine learning, AI, LLM