A Neural Network Tool to Organize Large Document Sets

Document clustering based on semantics is a fundamental method of helping users to search and browse in large cllections of documents. Recently a number of papers have reported the applications of self-organizing artificial neural networks in document clustering based on semantics. In particular Growing Neural Gas is a growing neural network that allows the user to reproduce the topological distribution of the inputs, but the structure obtained often has the same complexity as the input data structure; if the input space has more than three dimensions it is impossible to visualize or represent the GNG network as well as the input data structure. In this paper the authors propose a LBG modified network, called LBG-m, that can simplify the GNG structure in order to visualize and summarize it. The two algorithms constitute a tool for browsing large document sets and generating a set of semantic links between clusters of similar documents.