Preprocessed Conversational Mental Health Dataset for AI Model Training

Title: Preprocessed Dataset for Conversational AI and Mental Disorder Classification Using GPT-3.5 and DistilRoBERTa Description: This dataset supports the research paper titled "Conversational AI for Mental Health Diagnostics using GPT-3.5 and Distil-RoBERTa". It contains a cleaned, filtered, and semantically selected dataset constructed by combining two publicly available sources: 1. Mental Health Conversational Data (Elvis, 2022): · Kaggle Link · Used for training the GPT-3.5-based conversational model with mental health-related dialogues, FAQs, and therapeutic scenarios. 2. Mental Disorders Identification (Reddit) (Adha, 2022): · Kaggle Link · Contains ~700,000 Reddit posts labeled for mental disorders. Used to train a distilroberta-base classifier for four classes: Anxiety, Depression, Borderline Personality Disorder (BPD), and Bipolar Disorder. To ensure high-quality, semantically rich examples, we applied the following: · Text cleaning and filtering to remove noise, low-context replies, and troll content. · Semantic sample selection using Sentence-BERT (all-MiniLM-L6-v2) for embedding generation. · t-SNE-based dimensionality reduction followed by Euclidean distance-based selection of top 1000 representative samples per class. This Zenodo entry includes only the final processed dataset (CSV format), label descriptions, and supporting metadata. All source code and training pipelines are openly available on GitHub:https://github.com/diwakardkk/Conversational-AI License: CC-BY-4.0Keywords: conversational AI, mental health, NLP, Reddit, t-SNE, Sentence-BERT, classification