Interdisciplinarity Measurement Method of Scientific Research Papers based on Adaptive Feature Selection

[Purpose/Significance] Interdisciplinary research can creatively solve complex problems in natural environment and human society through knowledge integration and penetration. With the increase of interdisciplinary research results, the evaluation of interdisciplinarity becomes increasingly necessary. How to establish an effective method for interdisciplinarity measurement and achieve a comprehensive measurement of scientific research papers is an urgent problem to be solved. [Method/Process] Based on the above background, this study takes the data of scientific research papers as the analysis source, deconstructs the interdisciplinarity of scientific research papers from multiple dimensions, constructs the feature set of interdisciplinarity of scientific research papers, and on this basis proposes the method for measuring interdisciplinarity based on the adaptive method of machine learning, and conducts a comprehensive measurement of interdisciplinarity. This study has certain positive significance for researchers to understand the interdisciplinary papers in the field. The work process is as follows: First of all, the basic concepts of interdisciplinarity are sorted out and related concepts are discriminated, and the index of interdisciplinarity of different dimensions is analyzed. Based on the connotation and characteristics of interdisciplinary research, the characteristic index of interdisciplinarity of scientific research papers is extracted from three dimensions: subject attribute, knowledge network topology and knowledge integration text content. Secondly, an interdisciplinarity measurement method based on machine learning is constructed. By analyzing information gain and feature similarity of input indexes and data in feature sets, a feature selection calculation method based on adaptive feature selection is proposed, and the accuracy of feature classification is maximized by machine learning classifier. At the same time, the feature subset that can best express the interdisciplinary is selected based on the adaptive selection of the minimum number of features, and the selected adaptive feature set is used in the calculation of the interdisciplinary of the paper, and the results of the calculation of the original feature set are analyzed comprehensively. Finally, an empirical study was carried out in the field of plant nanobiotechnology to verify the effectiveness of the index system and adaptive feature selection listed above, identify and screen papers with high interdisciplinarity in the field, measure the interdisciplinarity of papers and identify key influencing factors based on the calculation of subject attributes, knowledge network topology and knowledge integration text content features. [Results/Conclusions] The main empirical results show that, among the subject attributes, the balance degree and the difference degree have a greater effect on the interdisciplinary evaluation. The overall effect of knowledge network topology structure features is satisfactory, the distribution breadth of knowledge integration text content features has a greater effect on interdisciplinary evaluation, and the calculation effect is further improved by fitness weighted summation of each feature. The results demonstrate that the adaptive feature selection proposed in this paper can effectively screen the interdisciplinary related feature indexes, improve the reliability of the results, and achieve a comprehensive and in-depth measurement of the interdisciplinary of scientific research papers. This measure method avoids the subjective defects that may occur in qualitative evaluation and the problems that different measure indicators may produce contradictory results. It provides a new idea and direction for interdisciplinary measurement.