A Comprehensive Review on Word Embedding Techniques

Word embeddings are a linguistic representation that can be used to gain insight into a person's mental process in order to train a computer to think like that person. Real numbers can be used as representations in the form of a vector. Word embeddings attempt to capture the meanings of words by depicting them randomly in an n-dimensional space. This research aims to survey the landscape of available word embedding techniques. Based on this survey, we can identify three widely used word embeddings: the “traditional,” “static,” and “contextualized” embeddings. BERT, a model based on two-way transformers that can be pre-trained and improved, makes contextualised word embeddings better. Combined word embedding and neural network (NN) models, has the potential to improve model accuracy in sentiment classification, text categorization, future phrase prediction, and additional natural language applications. Some issues and challenges are also discussed, as are the next steps for research towards better word representation.