Measuring healthcare Innovations through Human Assigned Approach and Man-Machine Derived Approach: A Comparative Analysis Using Published Patents of India

Aim

This study examines the effectiveness of the Latent Dirichlet Allocation (LDA) model in extracting thematic structures from healthcare patents and compares machine-generated topics with human-assigned International Patent Classification (IPC) codes. It also assesses whether using both patent titles and abstracts improves topic identification compared to titles alone.

Research Design and Methods

Healthcare-related patents published in India between 2000 and 2022 were retrieved from the WIPO PATENTSCOPE database. IPC classifications served as the benchmark for human-assigned categorization. LDA-based topic modeling was applied to patent titles, abstracts, and their combined text, and the resulting topics were compared with IPC classifications to assess alignment and thematic coverage.

Findings

IPC analysis identified key innovation areas, including medicinal preparations, organic active ingredients, and herbal drugs. LDA applied to titles highlighted themes such as crystalline pharmaceutical and herbal compositions, while abstract-based analysis revealed more detailed topics, including antiviral agents and rotavirus vaccine compositions. Although LDA effectively extracted latent topics, title-only analysis provided limited thematic depth.

Implications and Recommendations

Combining patent titles and abstracts significantly improves the accuracy and comprehensiveness of LDA-based topic modeling. While machine learning supports large-scale patent analysis, human expertise remains crucial for interpreting results and refining trend analysis. Hybrid analytical approaches are therefore recommended.

Contribution and Value Added

The study confirms the usefulness of machine learning for healthcare patent analysis in the Indian context. It adds methodological value by demonstrating the benefits of multi-field textual input and highlights the complementary roles of automated models and expert judgment in patent analytics.