An Integrated Human-in-the-Loop Patent Search Workflow: An Exploratory Case Study Combining Keyword, Similarity, Semantic, and Citation-Based Retrieval

Abstract

Patent retrieval is important for innovation management, R&D planning, and intellectual property strategy, yet conventional searches are hindered by terminology variability, synonymy, broad patent scopes, and classification limitations. This study presents an integrated, iterative workflow combining keyword-based retrieval, semantic analysis, document-similarity searches, citation-based refinement, and consensus-based manual relevance assessment. The workflow is illustrated through a single exploratory case starting from the same reference patent used by Moeller and Moehrle. After correction of duplicate entries, the final dataset contains 49 distinct patent identifiers: 19 were assigned to the high-relevance category, 20 to the related category, and 10 to the peripheral-relevance category. Six of the ten documents listed in the reference study were also retrieved; this 60% document-level overlap is descriptive and must not be interpreted as recall because the published list is not an exhaustive ground truth. The case illustrates how complementary retrieval channels, set intersections, iterative seed updating, and human decisions can be organized in a transparent workflow. It does not demonstrate superiority, general effectiveness, or equivalence between non-domain experts and expert searchers.