HLabelSOM: Automatic Labelling of Self Organising Maps toward Hierarchical Visualisation for Information Retrieval

The self organising map technique is an unsupervised neural network that is able to cluster high-dimensional data and to display the clustered data on a two-dimensional map. However, without the help of a visualisation technique or labels assigned by an expert, users without some prior knowledge will find it difficult to understand the clusters on the map. In this paper, we present the HLabelSOM method to automatically label the self organising map by utilising the features the nodes on the map hold after the training process. Users will be able to see the clusters on the labelled map since the neighbouring nodes have similar features and the labels themselves reveal the cluster boundaries based on the common features held by neighbouring nodes. Further, the HLabelSOM method produces several maps that can be utilised to create hierarchical visualisation for information retrieval. We demonstrate the applicability of the HLabelSOM method in mining medical documents from the Internet and visualising the information in hierarchical maps.