Feature-level sentiment analysis can retrieve the sentimental preferences for the features of products but cannot retrieve the causes of the preferences. Previous sentiment analysis methods used sentiment words to calculate the sentiment polarity for specific features but could not utilize neutral sentiment words, even when they constituted a large proportion of the sentiment words. Fault diagnosis can extract causes and determine the root cause by using factual information and the cause-effect relation, but is not used for sentiment data. For the retrieval of sentiment root causes, we propose a sentiment root cause analysis method for user preferences. We consider sentiment relations based on fuzzy formal concept analysis (FFCA) to extend hierarchical feature-level sentiment analysis. A hierarchical relation of neutral sentiment words and explicit causal relation based on causal conjunctions is utilized to retrieve the cross features of root causes. A sentiment root cause is determined from the extracted causes to explain the preference of a sentiment expression by using a fuzzy cognitive map with a relations method. We demonstrate a factual ontology and sentiment ontology based on a feature ontology for clothing products. We evaluated the proposed sentiment root cause analysis method and verified that it is improved as compared with term frequency-based methods and sentiment score analysis.
|Journal||Journal of Computing and Information Science in Engineering|
|Publication status||Published - 2016 Sep 1|
ASJC Scopus subject areas
- Computer Science Applications
- Computer Graphics and Computer-Aided Design
- Industrial and Manufacturing Engineering