METHODS AND APPARATUSES FOR EMBEDDING WORD CONSIDERING CONTEXTUAL AND MORPHOSYNTACTIC INFORMATION

摘要

The present invention relates to a word embedding method and apparatus in consideration of context information and morphological information of a word, and a word embedding method according to an embodiment of the present invention does not know out of vocabulary (OOV) in a sentence to be learned. Processing a sentence by replacing it with an unknown token of, inputting a character of a target word excluding the unregistered word from the processed sentence as an input of a context character model to be learned, Combining the surrounding context vectors with respect to the surrounding words of the target word in the sentence and setting the context text model as an initial state; And predicted embedding of the target word generated by connecting a forward hidden state and a backward hidden state calculated from the context text model and real embedding of the target word. ) To minimize the error between, and learning the context text model.