While depression is globally on the rise, the mental health sector struggles with handling the increased number of cases, especially since the pandemic. These circumstances have resulted in an increased interest in the e-mental health sector. The dataset is constituted of 67 857 posts from the most popular English-language online health forums between 15 February 2016 and 15 February 2019. The posts were first automatically labelled (biomedical vs. psy framing) via deep learning; second, the time series of framing types of recurring forum users were analysed; third, the clusters of biomedical and psy patterns were analysed; fourth, the discursive characteristics of each cluster were analysed with the help of topic modelling. Five ideal-typical patterns of forum socialization are described: the first and the second clusters express the developing of a ‘recovery helper’ role, either by opposing expert discourses or by identifying with the psy discourses; the third cluster expresses the acquiring of a substantively diffuse, uncertain role; the fourth and fifth clusters refer to a trajectory leading to the incorporating of a biomedically framed patient role, or a therapeutic psy subjectivity. Elements of data collection that potentially undermine representativeness: online forum users, open and public forums, keyword search. The trajectories identified in our study represent various phases of a general forum socialization process: newcomers (cluster 3); settled patient role (cluster 4) or psy subjectivity (cluster 5); recovery helpers (cluster 1 and 2).

In our study we assess the responsiveness of Hungarian local governments to requests for information by Roma and non-Roma clients, relying on a nationwide correspondence study. Our paper has both methodological and substantive relevance. The methodological novelty is that we treat discrimination as a classification problem and study to what extent emails written to Roma and non-Roma clients can be distinguished, which in turn serves as a metric of discrimination in general. We show that it is possible to detect discrimination in textual data in an automated way without human coding, and that machine learning (ML) may detect features of discrimination that human coders may not recognize. To the best of our knowledge, our study is the first attempt to assess discrimination using ML techniques. From a substantive point of view, our study focuses on linguistic features the algorithm detects behind the discrimination. Our models worked significantly better compared to random classification (the accuracy of the best of our models was 61%), confirming the differential treatment of Roma clients. The most important predictors showed that the answers sent to ostensibly Roma clients are not only shorter, but their tone is less polite and more reserved, supporting the idea of attention discrimination, in line with the results of Bartos et al. (2016). A higher level of attention discrimination is detectable against male senders, and in smaller settlements. Also, our results can be interpreted as digital discrimination in the sense in which Edelman and Luca (2014) use this term.