Identifying Psychological Theme Words from Emotion Annotated Interviews

摘要

Recent achievements in Natural Language Processing (NLP) and Psychology invoke the challenges to identify the insight of emotions. In the present study, we have identified different psychology related theme words while analyzing emotions on the interview data of ISEAR (International Survey of Emotion Antecedents and Reactions) research group. Primarily, we have developed a Graphical User Interface (GUI) to generate visual graphs for analyzing the impact of emotions with respect to different background, behavioral and physiological variables available in the ISEAR dataset. We have discussed some of the interesting results as observed from the generated visual graphs. On the other hand, different text clusters are identified from the interview statements by selecting individual as well as different combinations of the variables. Such textual clusters are used not only for retrieving the psychological theme words but also to classify the theme words into their respective emotion classes. In order to retrieve the psychological theme words from the text clusters, we have developed a rule based baseline system considering unigram based keyword spotting technique. The system has been evaluated based on a Top-n ranking strategy (where n=10, 20 or 30 most frequent theme words). Overall, the system achieves the average F-Scores of .42, .32, .36, .42, .35, .40 and .40 in identifying theme words with respect to Joy, Anger, Disgust, Fear, Guilt, Sadness and Shame emotion classes, respectively.