--- title: OnCoCo Message Classifier emoji: 🧠 colorFrom: blue colorTo: indigo sdk: gradio sdk_version: 6.10.0 app_file: app.py pinned: false license: cc-by-4.0 short_description: Classify psychosocial online counseling messages tags: - text-classification - counseling - mental-health - multilingual - german - english --- # OnCoCo Message Classifier This Space provides an interactive demo for classifying individual messages from psychosocial online counseling conversations into fine-grained categories. ## What it does Given a single counseling message — prefixed with the speaker role (`Counselor:`, `Client:`, `Berater:`, or `Klient:`) — the model predicts the most likely communication act categories and their confidence scores. Categories cover 40 counselor acts (e.g. empathic reflection, problem exploration, motivational interviewing techniques) and 28 client acts (e.g. expressing emotions, describing problems, requesting information). ## Model [th-nuernberg/xlm-roberta-base-online-counseling-oncoco](https://huggingface.co/th-nuernberg/xlm-roberta-base-online-counseling-oncoco) A fine-tuned XLM-RoBERTa model trained on the OnCoCo dataset using 5-fold stratified cross-validation. ## Dataset [th-nuernberg/OnCoCoV1](https://huggingface.co/datasets/th-nuernberg/OnCoCoV1) OnCoCo (Online Counseling Conversations) is a bilingual (German/English) corpus of annotated psychosocial online counseling messages. Topics include mental health, relationships, substance use, and financial problems. ## Usage Select one of the provided examples or enter a custom message. Prefix the message with the appropriate role marker: - `Counselor:` / `Berater:` for counselor messages - `Client:` / `Klient:` for client messages Use the Top-K dropdown to control how many top predictions are shown. ## Citation If you use this demo or the underlying dataset/model in your research, please cite the associated publications (see the dataset card for details). ## License CC BY 4.0 — Technische Hochschule Nürnberg, Jens Albrecht