A Hybrid Model for Preference Data
忠 今泉 · Institutional Repositories DataBase (IRDB) · 2011
Preference scores to n objects of N individuals is a popular data collected in Marketing, Behavior Science, etc. A vector model or an unfolding distance model have been used to analyze these type of data matrix. However, it is difficult to understand what attributes contribute on preference evaluation using these continuous mapping models as the decomposition of data is not unique. The overlapping cluster models and methods such as ADCLUS (Shepard and Arabie, 1979) have interesting features to find the attributes in similarity data. So we propose a modified model of overlapping model, a hybrid model, to discover the hidden attributes of objects by putting a decomposition constraints. And we also show an application to real data set.