engleski

Content-based Recommender System for Textual Documents Written in Croatian

The paper describes a content-based recommender system that classifies textual documents written in Croatian. We describe how documents are pre- processed, including procedures of dimensionality reduction, selection of stop- words and creation of document-term matrix. For the text classification, a combination of v- fold cross validation and k - nearest neighbours (kNN) methods is used. This way, the ‘optimal’ value of k is firstly analyzed, and the results of v-fold cross validation are applied for the selection of value k. Results are given in the form of classification error analysis.

text mining; recommender system; k-nearest neighbour; content-based classification; document-term matrix.

nije evidentirano