HopeDetect: a multicomponent deep learning framework for hope detection in Kurdish language

Abstract This paper presents a novel deep learning framework for detecting expressions of hope within textual content and introduces KurdHope, the first hope-annotated dataset for the Kurdish language. Our proposed model integrates several specialized components, including a hope-aware attention mechanism, a temporal hope analyzer, and an emotion-aware processing system, all built upon a pre-trained BERT architecture. Moreover, we developed KurdHope, a large-scale dataset containing over 90 000 news headlines collected from popular Kurdish websites. These headlines were manually annotated by three independent annotators, who classified each one as either expressing hope or not expressing hope. Using this novel dataset, we evaluated our framework against traditional machine learning approaches and state-of-the-art deep learning model. The proposed model demonstrated superior performance, achieving an F1 score of 95.2%, an accuracy of 95.4%, a precision of 95.6%, and a recall of 94.8%. The proposed model significantly outperformed other baseline models, including Multilingual-BERT (which achieved a 93.5% F1 score) and traditional approaches like Support Vector Machine (with a 93.2% F1 score). Our results highlight the effectiveness of combining multiple specialized components to detect hope in text, particularly in capturing the complex interplay between linguistic, temporal, and emotional aspects of hopeful expressions.