Selforganizing Classification on the Reuters News Corpus

Selforganizing classification on the Reuters news corpus Stefan Wermter The Informatics Centre School of CET University of Sunderland St. Peter’s Way, Sunderland SR6 0DD United Kingdom Stefan.wermter@sunderland.ac.uk Chihli Hung1 The Informatics Centre School of CET University of Sunderland St. Peter’s Way, Sunderland SR6 0DD United Kingdom Chihli.hung@sunderland.ac.uk 1 Hung is a lecturer of De Lin Institute of Technology as well. Abstract In this paper we propose an integration of a selforganizing map and semantic networks from WordNet for a text classification task using the new Reuters news corpus. This neural model is based on significance vectors and benefits from the presentation of document clusters. The Hypernym relation in WordNet supplements the neural model in classification. We also analyse the relationships of news headlines and their contents of the new Reuters corpus by a series of experiments. This hybrid approach of neural selforganization and symbolic hypernym relationships is successful to achieve good classification rates on 100,000 full-text news articles. These results demonstrate that this approach can scale up to a large real-world task and show a lot of potential for text classification. Introduction Text classification is the categorization of documents with respect to a set of predefined categories. Traditional neural techniques for classification problems cannot present their results easily without adding extra modules but selforganizing memory networks (SOM) are capable of combining topological presentation with neural learning. We extract suitable relations from WordNet to present a semantic map of news articles and show that these relations can complement neural techniques in text categorization. This integration of SOM and WordNet is proposed to deal with the text classification of news articles. The remainder of this paper is organised as follows. In Section 1, we give a brief review of SOM. Section 2 is dedicated to a description of methods of dimensionality reduction. In section 3 of our hybrid neural approach, the new version of the Reuters corpus and the results of our experiments are presented. 1 Selforganising Memory for Learning Classification According to the theory and the organisation of biological systems, neurons with similar functions are placed together. Based on this idea, Kohonen proposed SOM (Kohonen 1982). SOM, based on an unsupervised learning principle, can map a multi-dimensional dataset into a low-dimensional space, usually 2-dimensional. SOM learns to place similar data on topologically close areas on the map. Therefore, people can choose the relevant clusters of documents on the map to get relevant documents. However, it is impossible for one map to encompass the continuously growing data source. In such cases, the categories are often arranged in a hierarchy or an adaptive structure, e.g. Incremental Grid Growing model (Blackmore and Miikkulainen 1993), Growing Cell Structures (Fritzke 1993), Hierarchical SOM (Wan and Fraser 1994), and Adaptive Coordinates (Rauber 1996; Merkl and Rauber 1997). Presentation and explanation are a possibly weakness for most ANN models to text classification. The robustness of the SOM algorithm and its appealing visualization effects make it a prime candidate in text classification (Lin et al. 1991; Ritter and Kohonen …

Content truncated to 500 words