Disentangling patent quality: using a large language model for a systematic literature review

Abstract Assessing patent quality has long been the subject of research interest due to interchangeable terminology, overlapping indicators, and diverse perspectives. To address these challenges, this study presents a comprehensive framework for assessing patent quality, that draws on stakeholder theory and adopts a multidimensional perspective encompassing economic, legal, and technological dimensions, each of which is clearly defined within the study. Using the capabilities of the large language model GPT-4, a systematic literature review was conducted, analyzing an initial sample of 5141 scientific articles and selecting 762 as relevant. From these selected articles, 985 distinct indicators for assessing patent quality were identified and classifed in accordance with the dimensions of patent quality. The findings reveal that forward citations, family size, and the number of claims are among the most frequently used indicators, highlighting a predominant focus on technological quality in nearly two-thirds of the literature. In addition, the study highlights several challenges in patent quality assessment, such as poor research reproducibility due to inconsistent definitions and applications of indicators such as family size. In response, eight research propositions are proposed, emphasizing the critical evaluation of indicators, the application of sophisticated methods, and the quantification of complex metrics. As a contribution to management and scholarship, this research underscores the complexity of patent quality assessment and provides a structured framework for future studies, emphasizing the importance of a multidimensional perspective. It also illustrates the transformative potential of large language models in enhancing systematic literature reviews, setting a new standard for future research.