Seminár Ústavu informatiky


Replicator and ART2 Neural Networks in Text Outliers Identifications

Mgr. Asmaa Mostafa Mahmoud Salem
Ústav informatiky, PF UPJŠ

17. mája 2017 (streda) o 12:45
SJ2S07 (3.16T), PF UPJŠ


The intuitive definition of an outlier should be: an observation which deviates so much from other observations as to arose suspicions that it was generated by a different mechanism (D. M. Hawkins, 1980). Two models of neural networks (1) Replicator Neural Networks and (2) neural networks based on Adaptive Resonance Theory were used in our work to identify outlier sentences in texts The results of both neural networks declare that they give very good answers for manipulated texts and quite good answers for original texts. In the manipulated texts, the insert sentences were covered as outliers. In the text analysis we used English and Arabic texts from recommended benchmarks.