Eureka! Institutional Repository

This dissertation researches and compares patent retrieval architectures using a combination of traditional first-stage retrieval algorithms and second-stage deep learning techniques. The author conducts experiments using various combinations of tools such as BERT models, the Pyserini indexing software, and the DeepCT software tools, and evaluates the effectiveness of each retrieval architecture. The datasets used, the field selection process, and the algorithm selection procedures are documented, along with the scripts and software developed for the experiments. The paper aims to determine the best techniques and methodologies for efficient and effective patent retrieval. The author concludes with their findings and proposes future research directions in this field. The study highlights the significance of information retrieval methods and their applications in everyday life in the 21st century. Furthermore, the paper provides an introduction to artificial intelligence, machine learning, and deep learning concepts, which are fundamental to understanding the thesis’s technical aspects.