Patent US 11,048,872

The present disclosure provides a word vector correction method using artificial intelligence technology. According to the present invention, a method of determining a word similarity for a specific document matrix including a plurality of N word vectors having M dimensions includes decomposing the specific document matrix into a word matrix represented in a D-dimensional word space using a D-dimensional weight matrix; calculating a similarity between the plurality of D-dimensional word vectors included in the word matrix obtained through the decomposition; and determining that words corresponding to word vectors satisfying a preset reference similarity value are similar to each other when the calculated similarity value satisfies the preset reference similarity value.