期刊名称：International Journal of Computer Information Systems and Industrial Management Applications
印刷版ISSN：2150-7988
电子版ISSN：2150-7988
出版年度：2011
卷号：3
页码：137-142
出版社：Machine Intelligence Research Labs (MIR Labs)
摘要：An important topic of Kansei engineering is to associate humans' impressions with physical features characterizing certain objects. Recently, such relationships havebeen applied to impression-based information systems that aim to retrieve/recommend objects suitable for users' impressions. However, such systems restrict the use of various expressions concerning each individual's image, because the selectableimpression words are fixed. To achieve information systems based on individuals' Kansei, it is important to provide a flexible human interface to deal with a greater diversity of impression words. In our previous study, we proposed a method for associating various impression words with physical features by introducing a new class of impression words called "meta-impressions." A meta-impression was defined as animpression whose relationship to physical features has beenrevealed in past studies. In the method, an M-I dictionary that describes the relationship between meta-impressions and various impression words was constructed using a text-mining technique. Using the dictionary, various impression words were translated into physical features through meta-impressions. However, theaccuracy of translation using the dictionary was not sufficient because the dictionary contained unnecessary or redundant relationships. In this study, we provide a new method for associating a large variety of impression words with physical features by implementing a function that filters out unrelated or redundant relationships in the M-I dictionary on the basis of the modification structures obtained by dependency parsing. Using the method, we try to construct a more accurate and exact M-Idictionary than that with the previous method. In the experiment, we show that our method improves the construction accuracy of the dictionary compared with our previous method.
