Una soggettazione automatica di letteratura grigia con algoritmi di rete neurale artificiale. Due esperimenti ICAS e ILC

The aim of this work is to create an automatic subject classification of grey literature documents using an artificial neural network. In particular, a software simulator of neural network with back-propagation learning scheme was used; training of the network was carried out on around 300 documents. The prototype developed follows the steps which were performed during the learning, the processing and the network querying phase. The analysis of the final tests provides targets to be referred to the percentage of document classification error for each subject. From this data it is possible to evince possible document-subject correlations and/or subject-subject correlations in order to construct a relational Database of the scientific documents available at the Institute of Computational Linguistics and at the Institute of Instrumental Analitical Chemistry.