Grant: Alliance Grants

Recent breakthroughs in AI have contributed to rapid advancements in the ability of computer to process natural languages (like English), with the dominant paradigm today being Large Language Models (LLMs) such as the popular ChatGPT. However, despite significant progress, many challenges still remain for deploying these systems broadly and safely. In particular, in many scenarios, users need efficient access to large amounts of information expressed in natural language, but LLMs cannot deal directly with multiple (possibly long) documents, because they are limited in their input size. In this research, we will develop NLP tools built on top of LLMs to support the user' complex analysis of multiple documents, with a focus on multiple reports and user generated content. For health care, we will work with the BC Cancer Registry to automate the processing of electronic clinical documents, such as pathology reports and radiology reports; while in a second project, we will work with the BC Mental Health and Substance Use Services to support the analysis of what a patient tells either to their psychiatrist or on social media about how they feel. On the business side, we will work with a global IT company, Salesforce, to develop tools to better understand the various topics and interests of customers from past interactions. Based on customers' generated content, we will also identify what customers like, dislike, and how their preferences change over time. This research will have huge economic and social benefits for Canada and beyond. Besides healthcare and customer support, our novel tools to understand and manipulate complex bodies of documents will help a diverse range of users, including journalists producing news and citizens making sense of those; as well as analysts in the energy and mining sectors, who by analyzing corporate reports will be better able to assess and prevent negative impacts on the environment and society. Overall, the proposed research has great potential to fuel these and many other NLP applications.