A Framework for Grouping and Comparing Feature Weighting Methods

X. Tong and M. Gu (Norway)


feature weight, nearest neighbor,similarity,weight space


Correct identification of feature relevance is a critical is sue in similarity-based problem solving systems, since it determines whether a proper experience can be retrieved to help solving the current problem. Many methods have been developed that give an 'important value', i.e. weight, to a certain feature to indicate its salience. In this paper, we propose a framework to distinguish different weighting methods which separately considers two aspects of weight ing methods: weight form and weight value determination. Weight form determines the capability to capture contex tual information to indicate feature relevance (i.e. deter mine weight scope), while weight value is the searching result in this weight scope. In addition, we briefly survey the current weighting methods under the structure of this framework.

Important Links:

Go Back