Vector-Based AI and Methodological Hybridization in Journalism and Media Research: A Structured Review and Exploratory Map of Analytical Profiles

Abstract

Recent journalism and media research increasingly uses embeddings, transformer-based models, and large language models, yet the methodological functions remain unevenly conceptualized. This article examines how vector-based and semantically oriented methods are incorporated into journalism and media studies and proposes a theory-informed framework of four recurrent methodological functions: semantic mapping, interpretive assistance, cross-scale articulation, and reflexive auditing. We conducted a structured review of English-language journal articles and review articles indexed in Scopus and Web of Science between 2022 and April 2026. The review combined database retrieval, conservative LLM-assisted metadata screening, full-text analytical extraction, expert manual validation, descriptive analysis, exploratory association tests, and an embedding-based map built from article-level analytical profiles. The final corpus comprised 43 articles. Empirical studies of news texts predominated, and semantic mapping was the most frequent primary methodological function. The profile-based map identified four moderately differentiated and overlapping communities centered on reception and circulation, semantic mapping of news content, affective and evaluative discourse analysis, and methodological infrastructure-building. Community membership aligned more strongly with media domain than with study type or primary methodological function, although these patterns are interpreted heuristically given the small corpus and low expected cell counts. The findings show that these methods are best understood as components of hybrid research designs combining semantic formalization, interpretation, validation, and scale-sensitive analysis.