Artificial Intelligence Tools for Business Applications: Objective Map of Science and Analysis of Texts

Abstract

Business is looking for technological and investment possibilities in research and development (R&D). Here the basic problems are to find R&D's results and/or teams for solving the professional tasks and for making investment. But business has no personal view on scientific problems. So business is seeking the objective tools for forecasting and evaluation of R&D prospects and results. Experts have own interests and require a lot of funding. R&D reflections are the texts. The modern methods of computer analysis of texts can do a lot of the experts' work for making it more objective and cheaper. The tools for search, systematization and ranking R&D's results and teams are computer analysis of texts and the map of science. The map of science is the distribution of the collection of texts of a scientific nature by the topics. The map of science is a way to navigate through the world of scientific publications and R&D's teams, a tool for identifying trends and assessing R&D directions. The usual ways for the map of science formation use bibliometric/scientometric data, general probability models of the texts, expert's opinion or artificial intelligence (AI) models and methods based on the thesaurus or on the ontology of the subject domains. The interests of business are not in line with the orientation on a priori established ideas about possible topics or there number for rapidly changing scientific fields. Precisely these fields are of the greatest interest to business. On the basis of mathematical modeling of texts and large-scale text collections, an approach is proposed for the computational formation of the adaptive dynamic map of science that does not use a priori classification schemes and data of the scientific publications' citation. Topics (thematic groups), their number and the distribution of texts over the topics are determined computationally without experts' involvement. Examples of the maps of science for various collections of scientific publications are given. The original method is proposed for checking the adequacy of the text models and the map of science. The method uses the categorization of articles and their abstracts as the separate objects on the basis of computationally generated map (its topics). The results of the large-scale experiment confirmed the high efficiency of the proposed mathematical modeling of texts and text collections. The possibilities of practical use of the map of science for business applications are considered.