Emotion-Aware Semantic Document Retrieval Using Sentence-BERT and Fine-Grained Emotion Classification

In the last several years, the spread of digital text in the present time has created a need for intelligence in retrieval systems that transcends lexical similarity and into semantic or emotive understanding. A new framework in the area of emotion-aware semantic document retrieval is presented in this respect. Sentence-BERT (S-BERT) embeddings are used in this framework to enable fine-grained emotional categorization. Though it might seem that models like CNN, RNN, LSTM, and CNN-LSTM are the most efficient for text modeling, these models seem to consistently miss the mark in accurately represent the deep contextual semantics produced from longer texts and the emotional subtleties they contain. Importantly, they meant to enhance retrieval performance by harmonizing the semantic similarity and emotional tone of texts since they use the transformer-based BERT and its sentence-level extension S-BERT. The system collects affect-based processing outcomes and scores them to assess their significance using emotion classifiers trained on exact class-label emotion categories. A comparative evaluation on benchmark emotion-rich corpora employing S-BERT, CNN, RNN, LSTM, CNN-LSTM, and vanilla BERT models is used to assess the efficacy of semantic retrieval and emotion recognition. This assessment is done to verify this efficacy. Conversely, in sentiment-sensitive sectors like healthcare, education, and social networks, fine-grained emotion categorization enhances user relevance experience. The findings of the studies show that S-BERT outperforms the traditional sequence models formerly employed in semantic retrieval operations.