In the context of word embeddings, what is the core idea behind 'Word2Vec'?

1. ক
   Representing words as high-dimensional sparse vectors
   
2. খ
   Predicting a word based on its context or vice versa using shallow neural networks
   
3. গ
   Using rule-based grammar to define word relationships
   
4. ঘ
   Mapping words to a fixed set of predefined categories