Detecting Hope in Social Media Discourse Using Machine and Deep Learning Classifiers

Hope speech refers to messages that convey optimism, support, or expectations of a better future. With the increasing use of social media as a medium for self-expression, analysing such messages may provide meaningful insights into the emotional well-being of individuals. However, hope speech detection has received comparatively limited attention in social media discourse analysis when contrasted with tasks such as hate speech detection. This study addresses this gap by conducting both binary and multiclass classification of hope speech in two languages: (i) English and (ii) Spanish. In the binary classification task, the objective is to distinguish between hopeful and non-hopeful tweets, whereas the multiclass task further categorises content into five classes: (i) Generalised Hope, (ii) Realistic Hope, (iii) Unrealistic Hope, (iv) Sarcasm, and (v) No Hope. Six traditional machine learning algorithms and three deep learning and transformer-based architectures were evaluated. The experimental results indicate that transformer-based models outperform traditional approaches in both languages. For English, RoBERTa achieved the highest performance (binary: 82.25% weighted F1; multiclass: 72.49% weighted F1), while for Spanish, XLM-RoBERTa attained the best results (binary: 86.32% weighted F1; multiclass: 77.01% weighted F1). These findings suggest the effectiveness of transformer-based models for multilingual hope speech detection. Smart citations: https://scite.ai/reports/10.61467/2007.1558.2026.v17i1.1239