TEXT ANALYSIS BASED ON BAG NEIGHBORHOOD MODEL AND APPLICATION TO INCIDENT REPORTS

摘要

A model for text analysis based on a real bag is developed in which a novel concept of bag neighborhood is used which takes nearness of words in a text sequence is taken into account. A condition such that a bag neighborhood defines a kernel that has been used in support vector machines is given. As a result kernel-based methods of data analysis can be applied to this model. Application to a set of medical incident report is studied using an algorithm of medoid clustering and principal component analysis.