A Survey and Future Perspectives of Hybrid Deep Learning Models for Text Classification

Abstract

AbstractThe advances in deep learning (DL) models have proven to achieve outstanding results in text classification tasks. This success is due to DL models’ potential to reach high accuracy with less need for engineered features. Despite their popularity, DL models have their strengths and weaknesses in their learning capacity, depending on the task. Researchers in the recent past have proposed various hybrid models to compensate for these weaknesses. This study presents a performance analysis of hybrid DL models compared to stand-alone DL models on various text classification tasks. Various research articles published between 2015 and 2020 on text classification using hybrid DL models were selected from leading computer science and engineering journals and analyzed. The findings suggest that hybrid DL models can better capture a syntactic representation of text, extract multiple feature maps, and effectively improve text classification results. The study also presents an improved awareness of different hybrid DL architectures in the field of text classification.KeywordsText classificationCapsule networkDeep learningNatural language processingSentiment analysis