Regret Detection on Social Media using BERT-BiLSTM Model

Regret is a common and fundamental emotion in everyone life. It is defined as the feeling that arises when we realize or believe that our current circumstances could have been better if different decisions had been made. It is accompanied by negative emotions such as remorse, self-blame, and disappointment, often shared by many users who regularly express their emotional experiences on the social web. Detecting the actual regret emotion of a sentence is complex because of its multifaceted nature. This article introduced the following steps for detecting regret content from English text: (i) topic modeling and automatic labeling using average agreement between Latent Dirichlet Allocation (LDA) and the K-mean clustering method, (ii) using pre-trained BERT base uncased embedding and TFIDF word vectorization for feature representation enables contextual analysis and assesses the relevance of terms, and (iii) Implementing a Bi-LSTM model stacked with word embedding techniques, and incorporating a single dense layer for regret text classification. The experimental findings demonstrate that the proposed learning approach (BERT-BiLSTM) has superior performance in identifying regretful sentences compared to other models. We evaluate the performance of the proposed model by assessing its precision, recall, F1 score, and accuracy metrics.