A Comparative Study on Improving Word Embeddings Beyond Word2Vec and GloVe

NLP or Natural Language Processing in Machine Learning forms a subarea with linguistic roots that has applications in analyzing and predicting natural language data, namely speech and text. Deep neural networks are used in cutting-edge NLP, and the process includes several steps: data collection, preprocessing, language modeling, parsing, and prediction. This research will focus on the aspect of language modeling where words must be represented using embeddings. The study emphasizes improving post-embedding model performance, particularly for unsupervised models, although the whole word embeddings used by these models have remained relatively unchanged. Word2Vec, for example, employs an unsupervised shallow 2-layered network that has suited its function thus far, but making this network denser and adding supervised elements appears to be a viable means of increasing spatial-semantic embedding. A better representation early in the model pipeline will significantly help any computer understand the semantic similarity of the data it is examining, which will considerably enhance the performance of any model that employs the stated representation.