Brain Dump

Continuous Bag of Words

Tags
text-processing

An approach in CWR.

Train shallow feed forward neural network to predict missing target word w in middle of 2k+1 words.

For example, to build a vector for the word aardvark.

  • Find many documents with the word aardvark.
  • Replace aardvark with one of k-words before and after aardvark.
  • Train the network to detect the queried word was aardvark from the documents

I.E. Train the network to identify the search term in the list of documents most relevant to that search term.