In order to consider user’s emotion and feeling on human-computer interaction system, it is important to understand the user’s personal preference. However, it is difficult to grasp personal preference information completely because it may differ among people and it can be changed easily by obtained knowledge and experiences. Although there are some database about preference and evaluation, most of them deal with “general” preference information. Nevertheless, people can estimate whether the partner likes an object from his/her utterances. For example, when a person happily says “X won the championship,” we can estimate that he/she likes X. On the other hand, when a person gloomily says “X won the championship,” we can estimate that he/she does not like X. In this paper, we propose a method to estimate like-dislike polarity for an object in an utterance by using such heuristics on the basis of the case frame structure of the utterance and the speaker’s emotion. The heuristics are expressed by the Emotion Generating Calculations method (EGC). The proposed method is applied into the utterance when the speaker expresses pleasure or displeasure. In the experiment, the proposed method calculated like-dislike polarity of a word in the utterance by using the speaker’s emotion estimated by a participant and the calculated polarity was compared with the polarity manually estimated by the same participant. The precision and recall of emotion estimation process were 0.76 and 0.88, respectively.
View full abstract