Resumen
To achieve competitiveness in a rapidly changing science, it is important to follow the development of existing technologies and discover new and promising technologies. Firms need to develop a technology development strategy by predicting technology trends in order to gain a competitive advantage while using limited resources. On the other hand, nowadays the number of scientific articles, patents and other miscellaneous data is growing at a rapid pace, and it becomes impossible to stay up to date with everything that is published. However, despite all efforts, none of existing methodological and technological results are able to create models and methods for the holistic perception of heterogeneous information by a computing system ? scientific publications and patents, which is contained in open sources. At the same time, most of the existing studies are intended for the analysis and early detection of new technologies or monitoring trends in some specific technology industries, without considering the solution of the problem of predicting many different technology trends. In addition, the accuracy of the assessment of the proposed methods in existing studies is either rather low (the maximum metric F1 for assessing the accuracy of the forecast is ~ 74%), or is absent (the quality of the method has not been assessed). Thus, this article proposes a new method for analyzing and predicting technological trends based on the processing of heterogeneous data (scientific articles, patents) from open sources by developing an algorithm for extracting significant keywords and methods for creating co-occurrence matrices of elements (keywords, CPC codes).