Abstract:
Generic features for word representation have been useful in the eld of natural language processing. In fact, many techniques have been introduced lately, each with its bene ts and downsides. Here we analyze di erent word embedding techniques for the eld of biomedical named entities recognition. This study found evidence that choosing the right word representation is as important as creating the right model. We tested Word2Vec, GloVe, and FastText and we found that FastText gives the most satisfying results. Besides, this manuscript proves that each method works well but only when used with the right problem.
