Embedding Digital News Titles Using WordNet Knowledge and WSD over BERT Model

The transformer model examines embedding approaches sensitive to context variation in natural language processing. Sense-enhanced embeddings, where words are represented as vectors, involve context-specific meanings for accurate sense disambiguation for digital news headlines. The approach integrates lexical knowledge bases, such as WordNet, and word sense disambiguation with contextual models, such as bidirectional encoder representations from transformers, to precisely identify the semantic meanings of words in digital news headlines. This study uses an open-access news corpus that combines symbolic and neural approaches. This study employs preprocessing, tokenization, sense detection, and transformer-based contextual training with hyperparameters. The final results demonstrate a significant improvement in WSD with BERT compared to WordNet with BERT in natural language processing, which is very relevant in the case of online news. This method addresses the primary issue of detecting contextual ambiguity in digital news headlines. Also impacts other applications such as machine translation, sentiment analysis, question-answering systems, and summarization tasks for online journalism.