Deep Learning Approaches To Simile Detection: Insights From Bert and LSTM

A crucial component of natural language processing (NLP) is simile detection. NLP seeks to improve comprehension of figurative language, particularly similes which utilize the phrases “like” or “as” for comparisons, to evoke strong feelings and vivid imagery. Detecting similes is important as it enables more nuanced text analysis, improving applications such as sentiment analysis, literary analysis, and language translation. Our approach involves comparing LSTM networks and BERT models to distinguish simile sentences from non-simile sentences. The extensive experimentation revealed that the simpler custom BERT model outperformed the complex LSTM model in accuracy, recall, F1 scores and precision metrics by around 12%. The results validate the approach and highlight the capability of pre-trained transformer models in enhancing figurative language processing. The work provides a foundational step towards more advanced and comprehensive language understanding systems, contributing to the broader field of NLP and its applications.