This paper investigates the use of natural language processing (NLP) techniques to analyze individual framings of depression in online forums, aiming to classify large digital corpora according to various discourses on depression. The study highlights challenges encountered with traditional human annotation methods and the limitations of algorithmic NLP in sociological contexts. Initial attempts showed significant inter-annotator disagreement, leading to a shift towards intersubjective hermeneutics. While machine learning performed well in predicting biomedical and psychological framings, it struggled with sociological framings, suggesting that sociological discourse on depression may be less well-defined. The paper emphasizes the need for further empirical study and a deeper understanding of the role of human annotation in machine learning applications in social sciences.