Hybrid optimisation enabled hierarchical deep learning for Facebook sentiment review analysis

Sentiment analysis is an important study nowadays due to sites of social networking, in which online users convey their thoughts, feelings and expressions freely on a particular topic. In this work, Facebook sentiment analysis is done by hierarchical deep learning for text (HDLTex). Here, Facebook reviews are first processed for bidirectional encoder representations from transformers (BERT) tokenisation. Also, many features, like negation, all caps, question marks, elongated units, hashtags, emoticons, number of words, and sentences in reviews, punctuation, a bag of units, and numerical words are extracted. Further, sentiment classification is carried out by HDLTex which is trained by the proposed Tasmanian devil driving training optimisation (TDDTO). Moreover, TDDTO is a combination of driving training optimisation (DTO) as well as Tasmanian devil optimisation (TDO). Furthermore, the performance of the proposed TDDTO_HDLTex provides an improved performance regarding the evaluation metrics like precision, recall, and F1-sore of 92.6%, 95.8%, and 94.2%.