Five-Level Lossless Knowledge Graph Dataset of a Qualitative Methods Text Corpus

This dataset provides a five-level, fine-grained, and structurally normalized knowledge-graph representation of a qualitative methods text corpus (Research with Qualitative Data), treated purely as text data rather than as a bibliographic object. Each record corresponds to a node at one of five hierarchical levels—macro-section (level 1), meso-section (level 2), paragraph (level 3), sentence (level 4), and keyword/media snippet (level 5)—with explicit parent–child links (e.g., sentence → paragraph, paragraph → meso-section), forming a well-closed, acyclic tree structure. For all machine-readable content in the source PDF, the dataset decomposes the corpus into independent nodes while preserving page locators and section titles, so that any fragment of text can be traced back to its exact position in the original file. Keyword nodes are automatically extracted from sentences to enhance search, thematic mapping, and downstream modeling without altering or compressing the underlying text. For tables and images, the dataset stores captions, surrounding textual context, and row-level data_points where applicable, enabling full reconstruction of tabular and visual information at the text level. Under the assumption that “all machine-readable text in the PDF is the reference universe,” the collection achieves a practically lossless representation of the qualitative methods corpus and has been independently checked for level completeness, parent–child consistency, and content integrity, supporting its designation as a five-level, completely lossless text-based knowledge-graph dataset suitable for advanced qualitative methodology research, knowledge-graph engineering, and large-language-model retrieval and reasoning experiments.