Emotional Computation of Unfamiliar Word based on Emotional Orientation of Related Words

Inhalt:
Traditional emotional computing approaches use emotion dictionary to search an emotional score, which cannot compute the emotional score of an unfamiliar word that is not in the dictionary. For this issue, this paper proposes a method called Emotional Orientation of Related Words (EORW) to compute the emotional score of an unfamiliar word. Firstly, the Word Vector Model calculates the word vector of an unfamiliar word. Secondly, based on the spatial vector similarity, the Adjacent Neighbor Related Words Set (ANRW-Set) of the word is calculated by the close distance. Then, through the emotional tendency of ANRW-Set, two Mutually Exclusive Subsets (ME-Subset) of related words are divided. Finally, the dominant emotion group is obtained by weights of two subsets. The emotional score of the unfamiliar word is calculated by the fusion of words in the dominant group. The experiments show that the accuracy is basically maintained at about 80%.