Quantifying Patterns Of Researchers’ Scientific Activity Based On The Domain Knowledge Map

China has made significant achievements in science and technology,but there are still problems such as the overall effectiveness of the innovation system needs to be improved,the scientific research evaluation system needs to be ameliorated,and the scientific research cooperation model needs to be perfected.Exploring and grasping the laws of scientific research activities is an important prerequisite for promoting the reform of the science and technology system,building an open innovation ecosystem,and improving the overall level of science and technology.From the perspective of complexity,science can be understood as a complex adaptive system composed of three elements: research subject,knowledge object and symbolic carrier,which is constantly expanding and evolving.In this complex research innovation system,the research subject is mainly the researcher,the knowledge object refers to the disciplinary domain knowledge,and the common symbolic carrier is the academic paper.The collective exploration and creation activities of researchers in the disciplinary domain knowledge space drive the continuous development of the research innovation system.Then,what patterns exist in the knowledge inquiry behavior of researchers in the domain knowledge space? Current research has focused on the fields of philosophy of science,sociology of science,and scientometrics.Although relevant studies have explored the issues of topic selection behavior,research collaboration behavior,topic identification and academic results evaluation of interdisciplinary research from different disciplinary contexts,the systematic investigation of the patterns of knowledge exploration behavior of researchers and their mechanisms is not sufficient.In particular,the research on the social factors influenced in the process of topic shifting of researchers,the evaluation of the academic performance of cross-disciplinary collaboration patterns,and the structural basis of the efficiency of research collaboration systems under this research problem needs to be studied in depth.In recent years,the accessibility of large-scale scientific and technical digital literature resources,the rise of social computing science,and the development of complex network theory have provided important opportunities to deeply explore the complex interactions between subjects and researchers’ behavioral mechanisms in scientific innovation systems.Therefore,taking the physics discipline as an example,this study addresses the scientific innovation system and theoretically explores the topic shifting and collaboration behavior patterns of researchers based on quantifying the knowledge space from the perspective of the interaction between the space of researchers’ knowledge inquiry behavior and the domain knowledge space.This research work is carried out in the following three aspects:First,as a basis for quantitative analysis of researchers’ knowledge inquiry behaviors,this study proposes a domain knowledge map construction method that integrates textual content features and citation structure features.In the context of and the era of big data and large disciplines,it is increasingly difficult to pursue accuracy and completeness in the detection and identification of domain knowledge structures.At the same time,most current studies only use the idea of text content features or citation network structure features,and the accuracy of their results is affected by a combination of factors such as multiple meanings of words,missing texts,and highly cited literature destroying the domain topic structure.Therefore,this study proposes a research framework for domain knowledge detection based on deep graph representation learning and stream learning algorithms to construct a domain knowledge map of the subject area more efficiently and more clearly.The research framework also provides a complete,unified and measurable knowledge space with continuous differences for exploring the mechanism of researchers’ topic shifting behavior.Second,to understand the activity patterns of researchers in the knowledge space,this study investigates the dynamics of researchers’ topic shifting behaviors in the knowledge space.The topic selection behavior of researchers can be viewed as the knowledge inquiry behavior of cognitively limited individuals in a complex cognitive landscape,and is influenced by both individual free will and social factors.Based on the domain knowledge map,the present study depicts the topic shifting trajectories of researchers in the knowledge space,measures the mobility scale and distance of research groups in different subspace regions,and detects the conservative topic shifting behavior patterns of most researchers in the local knowledge space at the group level.Further,the results of simulation modeling reveals that the core factor that promotes researchers’ topic shifting behavior is the research hot spots formed by the number of people in the spatial region,and the factor that inhibits researchers’ topic shifting is the barrier constituted by the knowledge distance between subregions.In contrast,although there may be research opportunities between subfields that promote scientific breakthroughs,empirical evidence suggests that this is not a significant factor in attracting topic shifting among researchers.Finally,to recognize the more complex patterns of cross-community or cross-domain collaboration among researchers in modern science,this study first analyzes the association between researchers’ propensity to collaborate across subfields and their academic performance at the micro-scale.Further,at the macroscopic scale,the central core structure in the collaborative network of researchers is probed to understand the structural features and functional performance of the emergence of the effectiveness of complex research innovation systems.On the one hand,by measuring the propensity of scholars to collaborate across subfields,it is found that the propensity of researchers to collaborate across fields is consistent with the theoretical model of "efficient teamwork".Furthermore,a U-shaped correlation pattern between the propensity to collaborate across subfields and their academic performance is revealed,and the trade-offs between specialization and diverse collaborative exploration are verified using the recognition cycle,academic output,and impact indicators of researchers.On the other hand,the multiplicity of heterogeneous core structures emerging at the mesoscopic level of research collaboration networks is identified by combining deep reinforcement learning pre-training models with information on the pivotal roles of complex networks.The existence of multiple heterogeneous core structures reflects the spontaneous formation of "local centrality and global decentrality" in the scientific collaborative system,which makes the collaborative system economical in structure and efficient in function.At the same time,comparing the academic performance indexes of members of different meso-core structures,it is found that the "Heterophilic club" formed based on cross-subfields collaboration plays an important role in bridging and integrating different knowledge fields,and disruptive index of it is higher than that of the "Rich club" members.The "Diverse club",based on cross-disciplinary collaborative relationships,was found to play an important role in bridging different fields of knowledge,and its disruptive index was higher than the "Rich club" members.Established upon the main findings of the above work,the key conclusions of this study include: 1)proposing a method for detecting domain knowledge structure based on graph neural network algorithms;2)discovering the mechanism of group-level influence on researchers’ topic selection and transformation;3)detecting the existence of not only the core-structure constructed by the tendency of collaboration intensity,i.e.,"Rich club",in the co-authorship network composed of researchers’ collaborative relationships,but also a diverse meso-nuclear structure formed by the tendency of collaboration breadth.We further analyzed the impact of different "exploitation" or "exploration" strategies on researchers’ innovation outcomes through the association between collaborative relationships and researchers’ academic performance.Through exploring the topic shifting pattern of researchers in the knowledge space,analyzing the choices and trade-offs they face in their research activities,and investigating the collaboration and division of cognitive labor in the process of knowledge exploration,this study aims to deepen our understanding and knowledge of the intrinsic factors driving scientific development and the collective knowledge discovery pattern,so as to provide references for establishing a better national innovation system and formulating more scientific and empiricalbased research management and innovation management policies.