Research on Example-based Text Categorization

摘要

The goal of text categorization is the automatic classification of documents into predefined categories. Text categorization usually training text corpus to create a classifier by machine learning technology, then analyses and compares the features of unlabeled documents with that of the classes of the classifier to classify it into the most similar category. Some algorithms support this method, such as Nearest Neighbor, Naive Bayes, Support Vector Machine, etc. This method has some disadvantages, such as complicated algorithms, fewer numbers and lower levels of classes. This paper proposes a new method of text categorization from a new angle. It uses manual indexing experiences, coming from some large bibliographic databases, to construct an example base for automatic classification. Each record of the base is an indexing record, including cross concordance of class numbers and strings. It can be used to realize text categorization through computing the similarity between feature strings of unlabeled documents with each indexing examples. Empirical results prove that this method have many advantages, such as simpler computation, more numbers and deeper levers of classes. This paper will introduce its algorithm, method of construction of example base for classification, and performance of system at length.