Forecasting News Events Using the Theory of Self-similarity by Analysing the Spectra of Information Processes Derived from the Vector Representation of Text Documents

This paper presents a new methodology developed by the authors to forecast news events based on representing texts as vectors, subtracting information process spectrums from the vectors, and analysing these with the help of self-similarity theory. Spectrum extraction is performed using text processing by applying mathematical linguistics approaches (text markups, normalisation, commenting); the vectors obtained are then clustered according to theme groups and the time of the news appearance. By applying the Hurst self-similarity method to analyse information news process spectrums, their self-similarity features are analysed. The information processes are then classified into two different classes: self-similar and not self-similar. It is proven that if the self-similarity feature is present in the processes investigated, the Hurst self-similarity method (R/S analysis) and almost-periodic functions will enable us to discover the almost periods of repeatability of events, and this will further allow us to forecast their behaviour over time and predict new events.