We mention the process where an article is given as the input and a list of keywords is generated as the output. We nominate the classifier which corresponds to the sub-group which is similar as the given article, based on its content. A list of words is extracted by indexing the article, and each word is encoded into structured forms. The extracted words are classified by the nominated classifier into ‘keyword’ or ‘not’, and the words which are classified into the former are selected. The performance depends on the granularity of each sub-group; it should be optimized between the two factors: the amount of sample examples and the subgroup granularity.