Self-Organizing Maps Applied to Information Retrieval of Dissertations and Theses from BDTD-UFPE

This paper proposes the application of Self-Organizing Maps (SOM) in the construction of an Information Retrieval System (IRS) to the Digital Library of Theses and Dissertations at Federal University of Pernambuco (BDTD-UFPE). The hypothesis is that the trained SOM and its graphical representation can help the user to obtain a general view of topics discussed in the document collection and also to perceive the similarity among documents, topics and graduate programs' knowledge areas and subjects. We present the system architecture and implementation's issues and evaluate the hypothesis through experimentation. The unsupervised organization performance of the proposed system was measured in terms of text categorization effectiveness and visual inspection of categories' distribution over graphics about trained SOM. The experimental results show that the proposed system generates relevant document map and confirms the hypothesis.