Mining and tracking the evolution of topics in a collection of documents helps identify and understand trends and shifts over time. This approach has proven particularly useful in bibliometric analysis, revealing how research topics in a field gain or lose prominence and helping researchers stay ahead in emerging areas of interest. Various methods have been employed to demonstrate the evolution of academic topics extracted from published articles. However, many of these methods rely heavily on extensive labeled datasets and struggle to accurately extract multi-word topics, resulting in incomplete maps of topic evolution. In this paper, we propose an Academic Topic Learning and Mapping (ATLM) model, a two-phase approach designed to learn and map academic topic evolution. By integrating an n-gram algorithm with zero-shot classification, the ATLM can extract academic topics from articles without the need for labeled data. A similarity-based method is then employed to identify the evolutionary relationships of topics over time. The efficacy of the ATLM is demonstrated in the context of the Australian National Disability Insurance Scheme (NDIS), a pilot personalized disability service in Australia that provides funding to support people with disabilities. Since the inception of the NDIS in 2013, this study is the first to collect and illustrate the key topics in the NDIS literature and the evolution of these topics over the past decade. The results are valuable for researchers and policymakers of the NDIS to better understand the development of critical issues and to guide future research and policy decisions.