TOPIC MODELING IN MANAGEMENT RESEARCH: RENDERING NEW THEORY FROM TEXTUAL DATA

Management scholars are increasingly using topic modeling to analyze large collections of text, a technique borrowed from computer science that identifies latent themes (topics) based on word co-occurrence patterns. This paper conceptualizes topic modeling as a rendering process—an iterative, three-stage cycle that bridges the gap between raw textual data and theoretical contribution. The authors argue that by framing the method this way, researchers can avoid treating it as a black box and instead use it to inform and sharpen their interpretive theory work.

The authors break down the rendering process into three distinct phases:

1. Rendering Corpora: This involves selecting and trimming raw text data, such as patents, media reports, or social media posts. This stage requires careful attention to language, authoring, and document sources to ensure the data fits the research question.
2. Rendering Topics: In this phase, researchers apply algorithms (like Latent Dirichlet Allocation) to identify latent topics. The authors emphasize that this is not a purely automated process; it requires the researcher to balance quantitative metrics of fit (like perplexity or coherence) with qualitative validity (semantic meaningfulness).
3. Rendering Theoretical Artifacts: The final stage involves iterating between the identified topics and existing theory to create constructs, causal links, or mechanisms. This is where the machine-generated output is transformed into a contribution to management literature.

Topic modeling allows management researchers to explore phenomena that were previously intractable due to the sheer volume of text. The authors identify five key areas where this method has already advanced theory: detecting novelty and emergence, developing inductive classification systems, understanding online audiences and products, analyzing frames and social movements, and understanding cultural dynamics. By providing a structured framework for using these tools, the paper empowers scholars to move beyond simple word counts and toward a more nuanced, theory-driven analysis of language, meaning, and social structure.