<p>The detection of hope speech has become an important area of research in natural language processing with potential applications in digital well-being, online community building, and positive content moderation. Previous research has predominantly focused on binary classification of hope speech, which fails to reflect its emotional and rhetorical nuances. We present a multi-dimensional computational model for fine-grained classification of hope speech in online text by extending binary hope speech detection to include two other levels of classification: hope type and hope purpose. The HopeEDI corpus was augmented through sentiment-based classification and zero-shot classification, with human verification, to include labels for three levels of hope intensity (Faint, Optimistic, and Highly Optimistic) and six hope purposes (Personal Story, Motivation &amp; Positivity, Career &amp; Achievement, LGBTQ Support, Religious/Spiritual Support, and Social Justice-Oriented Expression). Five transformer-based models (ALBERT, RoBERTa, XLNet, XLM-R, and Multichannel BERT) and five traditional models were tested on binary classification and three multiclass tasks. Our experiments show that transformer-based models outperform classical machine learning models, with Multichannel BERT performing best (F1-score 0.98 and AUC 0.99). Explainability experiments conducted with LIME identified linguistic cues linked to different levels and intentions of hope, thereby enhancing interpretability of the classification results. The research also introduces HOSAC, a conceptual deployment model for leveraging hope speech analysis in content moderation, mental health, and community-building technologies.</p>