This paper proposes a linguistic-statistical method for an automatic extraction and identification relationships of key phrases in information flows in order to further identification narratives as a generalization of a set of such phrases. Means of detecting stable phrases are based on the concepts of machine learning, linguistic analysis and statistical calculations. Next, using modern methods of network analysis, the relationships of key phrases are researched, and their individual clusters that probably correspond to the narratives will be identified. A form of visual display of information flow in terms of key phrases and dates is proposed. This form is a Phraseme Diagram (Ph-Di), the cells of which are filled with numerical values corresponding to the frequencies of key phrases in relation to the dates on which they appear. This approach can be used to address the analysis and visualization of the distribution of narratives for any selected information flows in terms of issues of researcher interest and have a fairly significant time frame. Keywords: Visualization, Costs, Time series analysis, Supervised learning, Machine learning, Network analyzers, Search engines, Thematic Information Flows, Key Phrases, Identification of Key Phrases, Extraction of Key Phrases, Relationship of Key Phrases, Dynamics of Key Phrases, Network of Key Phrases