We present a method and an interactive software to assist in the creation of hypertext links in a text, whatever the topic of the text and the application domain may be. The method follows three steps: 1) preprocessing of the text for its segmentation and its "vectorization", 2) classification of segments of texts using ART, an unsupervised learning neural net, 3) research by genetic algorithm of the best associations between segments of text by using the classification provided in step 2 and the subjective choices expressed interactively by the user. Steps 1 and 2, carried out thanks to CONTERM software, are presented briefly. Step 3, implemented in PROGEN software, is developed and the results of a first experiment are presented. A discussion on the interest and the limits of the method are given in conclusion.
展开▼
