A technique for forming, clustering and visualizing so-called directed correlation networks is herein proposed. The links between nodes of such networks correspond to the values of cross-correlations between vectors - sets of parameters corresponding to these nodes modified in a certain way. To build network structures for each node (topic), vectors are formed - arrays of numbers corresponding to a certain time series. As an example, the article considers a time series generated by the Google Books Ngram Viewer service.
This approach, unlike the existing one, has advantages such as intuitive and realistic rules, the definition of the weight of nodes and links, a reliable mathematical basis for correlation analysis, an accounting of previously unused parameters of time series of publications corresponding to entities, allowing one to the group said entities according to their trends in time, and objectivity and relative simplicity. This technique can be based on data obtained, for example, from content monitoring systems, and can be used in analytical systems for various purposes in order to generalize a set of variables without explicit links between them.
Distributed in Computational Biology eJournal Vol 4, Issue 50, November 04, 2020.
Distributed in Machine Learning eJournal Vol 3, Issue 126, November 09, 2020