Project Title: AI Language Models to Generate Optimisation Models and Solution Approaches   
Student: Yisi Chen
Course: MS in Computer Science (Artificial Intelligence)  

Abstract: 
Linear optimization is an important medium to guide businesses and decision makers to make 
informed and beneficial choices that ultimately save costs and improve competitive advantage. 
The process of solving a linear optimization problem is fundamentally built around four key 
components: variables, constraint equations, parameters, and objective functions. Traditionally, 
the solution of these problems has relied heavily on manual analysis by experts in the field. 
This reliance on expertise can lead to significant challenges for beginners or newbies who 
are new to linear concepts, making it difficult for them to effectively master and apply these 
techniques. However, with the rapid advancement of artificial intelligence and the development 
of big data, part of the work of linear optimization can be identified and processed automatically 
by LLMs trained by the system. In order to increase the interest of beginners or newbies in 
linear optimization and let them have a better understanding of linear optimization, I designed a 
chatbot with a function about transportation optimization problems. The chatbot has the ability 
to analyze user input, solve user needs, provide solutions to user queries, and handle tasks 
related to linear optimization such as answering basic questions and actually solving linear 
optimization problems. To achieve this, chatbots utilize a range of LLM models, such as GPT models 
and BERT models. The chatbot uses a vector space model (VSM) to analyze user input. Token limitations 
for post-processing of natural language input can be a challenge, as this may limit the chatbot's 
ability to handle lengthy or complex queries. To overcome this problem, the strategy chosen by the 
chatbot is to divide user input into smaller segments and assign labels, then transform it through 
a two-stage approach, and finally solve the linear optimization problem. The test results show that 
the transportation optimization problem can be partially solved by the chatbot. In addition, it is 
user-friendly and easy to access. Despite these successes, there are still some areas that need 
further improvement, for example, chatbots currently miss recognition in named entity recognition, 
misrecognition, etc. But overall, chatbots have received high accuracy and reviews.