Human-robot interaction (HRI) is becoming more complex and difficult due to the growing number of capabilities demonstrated by socially interactive robots. From this perspective, it is necessary for us to have simple and general methods that enable robots to interact with human beings at an emotional level. Therefore, this paper suggests a method that can efficiently use the PAD emotion space to generate artificial emotions based on a categorization of emotions using cluster analysis. After clustering, emotional appraisals can be extracted according to PAD input vectors using results of the categorization then we can generate blended emotions and calculate an intensity of each emotion. We evaluate the categorization results by using the Davies-Bouldin index and the patterns of emotions that are generated through categorization results. Furthermore, the generated emotions are expressed by a physical mascot-type head robot using our emotion expression model. We will also show how differently emotions are generated according to whether optimization is used or not.