نوع مقاله : علمی-پژوهشی
نویسندگان
1 گروه مهندسی کامپیوتر- دانشکده مهندسی برق و کامپیوتر- دانشگاه تبریز- تبریز- ایران
2 دانشکده مهندسی برق و کامپیوتر- دانشگاه تبریز- تبریز - ایران
3 گروه مهندسی مخابرات- دانشکده مهندسی برق و کامپیوتر- دانشگاه تبریز- تبریز - ایران
چکیده
کلیدواژهها
عنوان مقاله [English]
نویسندگان [English]
Extracting sentiment from textual data is crucial for understanding public opinion and guiding strategic decisions. We introduce a hybrid deep-learning pipeline that combines convolutional feature detectors, bidirectional recurrent units, and a custom attention mechanism. First, convolutional layers with pooling condense local n-gram patterns into compact feature maps. These maps are fed into a bidirectional LSTM network that captures sequence information in both forward and reverse directions. A specialized attention module then assigns relevance scores to individual tokens, sharpening the final classification. Evaluations on widely used sentiment benchmarks show that our method outperforms leading models in terms of accuracy as well as requires fewer computational resources, making it a practical solution for scalable emotion detection in text.
کلیدواژهها [English]
)