Methodology for data corpus construction in digital platform research

The article substantiates a methodology for constructing a data corpus in digital platform research using YouTube metadata as an empirical case. It distinguishes between a primary dataset and a corpus understood as a methodologically organized set of observational units selected through transparent thematic, genre, format, and functional criteria. A multilevel procedure is proposed for moving from the technical collection of digital traces to an analytically usable corpus. It includes consecutive analysis of channels, channel sections and playlists, and the metadata of individual videos. For each level, the article specifies indicators including titles and descriptions, statistics, publication formats, thematic relevance, functional orientation, author categorization, and engagement metrics. Particular attention is paid to a thematic relevance index and a cross-level consistency index, which make it possible to rank materials, establish inclusion thresholds, identify ambiguous cases, and document the logic of selection. The article also discusses the use of the YouTube Data API v3, Python, spreadsheet and relational data environments, and artificial intelligence services for metadata classification. AI is treated as an auxiliary tool for standardized coding rather than a substitute for researcher judgement, and its outputs require expert validation. The main limitations are associated with incomplete metadata, platform dynamism, dependence on API policies and quotas, algorithmically structured content visibility, and the risks of machine classification. The study concludes that the proposed approach strengthens the transparency, reproducibility, and validity of digital content selection and provides a controlled empirical basis for further sociological and interdisciplinary analysis.