Forecasting news events based on the model accounting for self-organisation and memory

The paper introduces a model of news events forecasting in news feeds developed based on stochastic dynamics of changes in the structure of the news clusters with memory of previous states and self-organization of their structure. Events forecasting is performed based on their textual description, text vectorization and calculating the cosines of the angles between the vector of the given text and centroids of various semantic clusters in the information system (IS). While creating the model, the probabilistic schemes of transitions from state to state at the IS are constructed, accounting for a set of previous steps. Based on the suggested approach, a non-linear differential equation of the second order is derived and the boundary problem for forecasting news events is formulated and resolved. This allows to obtain a theoretical dependency on time of the probability density distribution function of non-stationary time-series parameters describing the IS evolution. Experimental test of the developed model proved it to be reliable and adequate.