Developing Japanese WordNet Affect for Analyzing Emotions

This paper reports the development of Japanese WordNet Affect from the English WordNet Affect lists with the help of English SentiWordNet and Japanese WordNet. Expanding the available synsets of the English WordNet Affect using SentiWordNet, we have performed the translation of the expanded lists into Japanese based on the synsetIDs in the Japanese WordNet. A baseline system for emotion analysis of Japanese sentences has been developed based on the Japanese WordNet Affect. The incorporation of morphology improves the performance of the system. Overall, the system achieves average precision, recall and F-scores of 32.76%, 53% and 40.49% respectively on 89 sentences of the Japanese judgment corpus and 83.52%, 49.58% and 62.22% on 1000 translated Japanese sentences of the SemEval 2007 affect sensing test corpus. Different experimental outcomes and morphological analysis suggest that irrespective of the google translation error, the performance of the system could be improved by enhancing the Japanese WordNet Affect in terms of coverage.