Zero-Shot Emotion Detection for Semi-Supervised Sentiment Analysis Using Sentence Transformers and Ensemble Learning

We live in a digitized era where our daily life depends on using online resources. Businesses consider the opinions of their customers, while people rely reviews/comments other users before buying specific products or services. These are usually provided non-normative natural language within different contexts and domains (in social media, forums, news, blogs, etc.). Sentiment classification plays an important role analyzing such texts collected from by assigning positive, negative, sometimes neutral sentiment values to each them. Moreover, these typically contain many expressed hidden emotions (such as happiness, sadness, etc.) that could contribute significantly identifying sentiments. address emotion detection problem part analysis task propose two-stage methodology. The first stage is unsupervised zero-shot learning model based sentence transformer returning probabilities for subsets 34 (anger, disgust, fear, joy, admiration, affection, anguish, caution, confusion, desire, disappointment, attraction, envy, excitement, grief, hope, horror, love, loneliness, pleasure, generosity, rage, relief, satisfaction, sorrow, wonder, sympathy, shame, terror, panic). output used input second stage, which trains machine classifier labels supervised manner ensemble learning. proposed hybrid semi-supervised method achieves highest accuracy 87.3% English SemEval 2017 dataset.