Computer analysis of the emotional modality of 20 million publications in PUBMED database indicates ways to increase the effectiveness of pharmacotherapy by identifying pseudoscientific publications aimed at negative emotional "pumping" of doctors

Abstract

The search for original publications on fundamental and clinical medicine that would produce results of the highest scientific quality represents an urgent need for every medical researcher. Such publications are essential, in particular, for the development of reliable treatment standards. The Englishlanguage resources PUBMED and EMBASE are essential to help in solving this problem. However, there is an obvious problem in assessing the quality of the studies found. The paper formulates a method for analyzing the texts of biomedical publications, which is based on an algorithmic assessment of the emotional modality of medical texts (so-called sentiment analysis). The use of the topological theory of data analysis made it possible to develop a set of high-precision algorithms for identifying 16 types of sentiments (manipulative turns of speech, research without positive results, propaganda, falsification of results, negative personal attitude, aggressiveness of the text, negative emotional background, etc.). On the basis of the developed algorithms, a point scale for assessing the sentiment quality of research was obtained, which we called the "β-score": the higher the β-score, the less the evaluated text contains manipulative language constructions. As a result, the ANTIFAKE system (http://antifake-news.ru) was developed to analyze the sentiment-quality of Englishlanguage scientific texts. An analysis of ~ 20 million abstracts from PUBMED showed that publications with low sentiment quality (β-score <0, that is, that the prevalence of manipulative constructions over meaningful ones) is only 19 %. In the overwhelming majority of thematic headings (27,090 out of 27,840 headings of the MESH system PUBMED), a positive dynamics of sentiment quality of the texts of publications is shown by years). At the same time, as a result of the study, 249 headings were identified with sharply negative dynamics of sentiment quality and with a pronounced increase in manipulative sentiments characteristic of the "yellow" English-language press. These headings include tens of thousands of publications in peer-reviewed journals, which are aimed at (1) legalizing ethically unacceptable practices (euthanasia, perversions, so-called "population control", etc.), (2) discrediting psychiatry as a science, (3) media the war against micronutrients and (4) discrediting evidence-based medicine under the guise of developing the so-called "international standards of evidence-based medicine". In general, the developed system of artificial intelligence allows researchers to filter out pseudoscientific publications, the text of which is overloaded with emotional manipulation and which are published under the guise of "evidence-based standards".

Keywords

• pharmacoinformatics
• evidence-based medicine
• data mining
• algorithms for evaluating the quality of publications