Research On Tree-based Sentiment Dictionary Based On Feature Mining In Affective Computing

Affective computing is a new study area in computing technology.A lot of research has been done in this area.Its purpose is to allow computers to understand sentiment as well as humans.At present,the most commonly used method of affective computing is based on sentiment dictionary.But these commonly used sentiment dictionaries are onedimensional.They simply store sentiment words and sentiment tendencies in a list.However,the same sentiment word may have different sentiment tendencies under different features or item classifications.This is a problem that commonly used sentiment dictionaries can't solve.In order to solve this problem,this paper proposes a tree-based sentiment dictionary.In the tree structure,child nodes under the same parent node have similarities,and with deepening of hierarchy and refinement of tree branches,the similarities become larger and larger;the similarities between child nodes under different parent nodes are very small,and with reduction of common ancestor nodes,the similarities decrease sharply.This paper uses this feature to store item classifications and features in a tree structure.And then,sentiment words are added to the tree to form a tree-based sentiment dictionary.In this way,each sentiment word in this tree-based sentiment dictionary corresponds to a feature or an item,which solves the problem that the same word may have different sentiment tendencies under different items or features.Because each child node in a tree-based dictionary is a feature or a sub-classification of its parent node,parent node contains child node,each node transfers its sentiment words to its parent node,which solves the problem that sentiment words can't be searched under corresponding feature or items in the dictionary.In this paper,a tree-based sentiment dictionary and a sentiment lexicon are constructed by using Bayes' theorem.Compared with sentiment lexicon,results of affective computing based on this tree-based sentiment dictionary show an average increase of 9.78% in accuracy.Based on positive texts,results show an average increase of 10.40% in precision,an average increase of 4.78% in recall rate and an average increase of 7.72% in F value.Based on negative texts,results show an average increase of 4.87% in precision,an average increase of 20.84% in recall rate and an average increase of 12.99% in F value.The result of affective computing based on tree-based sentiment dictionary is better than that of sentiment lexicon.The main innovation of this paper is to propose a tree-based sentiment dictionary to solve the problem that the same word may have different sentiment tendencies under different items or features.This paper focuses on the construction of tree-based sentiment dictionary and affective computing based on tree-based sentiment dictionary.In the part of construction of tree-based sentiment dictionary,this paper first introduces how to construct item tree by using the relationship among item classifications,then introduces how to mine features and extract sentiment words by using syntactic parsing and association rules,and finally constructs a tree-based sentiment dictionary.In the part of affective computing based on tree-based sentiment dictionary,this paper first introduces the method of searching sentiment words in tree-based sentiment dictionary,then introduces how to calculate sentiment tendencies of sentiment words,and finally introduces the method of calculating sentiment tendency of text.