Leximancer Thematic Analysis Pipeline

The qualitative component of the PhD involved 60 semi-structured interviews with remote workers across Australia. Manually coding 60 interviews is tractable but slow. Leximancer provides a computational alternative: it generates concept maps from text corpora by identifying co-occurring concepts and their relative importance, which gives you a quantitative view of what themes are present and how they relate before you begin detailed manual coding.

The pipeline automates the pre-processing work that would otherwise be repetitive and error-prone: transcript cleaning (removing interviewer prompts, filler words, PII), speaker diarisation (separating participant responses from interviewer questions), and stopword tuning (domain-specific terms that are frequent but not conceptually meaningful need to be suppressed). The output of the pipeline feeds directly into Leximancer without manual steps.

The most valuable output was the concept co-occurrence matrix, which I used to triangulate against the SEM constructs before running CFA. If Leximancer identified "liminal space" and "disconnection" as co-occurring concepts in the interview data, but the survey items for Liminal Spaces didn't load onto a coherent factor, that's a signal that the operationalisation needed revision. Using qualitative and quantitative approaches in parallel created a feedback loop that strengthened both.