Basic Difference between Word Embeddings and Word Vectors
If you are Data Scientist or especially a Natural Language Processing expert then you should have come across these two terms while learning from basic to advance. Well, in this post I will discuss the major difference between these two concepts, although they might look similar with respect to the use case they are being applied for.
- Well, the first and most prominent difference is that the word vector uses a one-hot encoding sort of concept to create vectors for different words, whereas word embeddings use sophisticated algorithms to transform a vector for words (word2Vec)
- In the case of word vectors, vector size is directly proportional to the vocabulary size which makes it high-dimensional vectors, word embeddings on the other hand have a fixed vector size of low-dimension (e.g. 100, 300, 720, etc.)
- In most cases, word vectors contain occurrences (or co-occurrences) level information for different words (i.e. discrete numbers or counts). Word embeddings contain contextual and semantic level information (i.e. continuous real numbers)
- Word vectors can be used where there is a use case to give importance to keywords or need to find an exact match. Whereas, word embeddings should be used where the context or semantic similarity has a high priority (e.g. king and queen should have similar vectors)
I hope that this short explanation was helpful :)
System or Process to achieve a specific goal is more important than the goal itself!