Abstract:

The traditional feature selection algorithms usually select features distinguishing the different types of documents by the evaluation functions. However, these methods take the separate word as unit to establish a vector space model. The important words in the documents and the relationship between words are  not realized. In allusion to the disadvantages mentioned above, a new feature selection algorithm based on the relationship between words is presented. This algorithm considers key words, mines words’ association and checks these association rules by a correlation analysis to produce a feature space which closely relates to the category attributes. The experiment indicates that this method is better to express the semantic content of the documents and has a good categorization result.

Key words: relationship between words;feature selection;association rule;text categorization