Tree-based Sentiment Dictionary for Affective Computing: a New Approach

摘要

Affective computing is an emerging academic study area in computing technologies. Its purpose is to allow computers to understand emotions as humans do. At present, for affective computing with textual signals, the most common approach is sentiment lexicon approach. The majority of sentiment lexicons are stored in list of words with positivity and negativity. However, the one word might have various sentiment tendencies in various context. This problem can't be solved by these traditional sentiment lexicons which refers to list-based sentiment dictionary in domain of affective computing. In order to solve this problem, this paper presents a new approach that is a tree-based sentiment dictionary. In the tree structure, the similarities between sibling nodes under the same parent node is great but sibling nodes under different parent nodes is small. This paper uses these features to store the item classification in a tree structure and adds the features and sentiment words of the item which are extracted by using syntactic analysis and association rules to the tree structure to form a tree-based sentiment dictionary. In this way, we solve the problem that the one word may have various sentiment tendencies in various context and the problem of finding no sentiment words under the item classification in the dictionary. Comparing with sentiment lexicons, the tree-based sentiment dictionary outperforms for affective computing in the criteria, such as precision, recall, F-measure, etc.