Scientific experiment evolution analysis method and device based on knowledge graph

Abstract

The invention discloses a scientific experiment evolution analysis method and a scientific experiment evolution analysis device based on a knowledge graph, which are applied to the technical field of computer software, and are used for carrying out vectorization representation on a scientific experiment according to a constructed scientific experiment knowledge graph to obtain a scientific experiment after vectorization representation. Secondly, performing clustering analysis on the scientific experiment subjected to vectorization expression by adopting a community discovery algorithm to obtain a clustering result, modeling the clustering result, constructing a scientific experiment evolution graph, and finally performing point and surface analysis on the scientific experiment evolution graph to obtain a scientific experiment evolution result. Quantitative calculation of the evolution process can be realized, subjectivity of evolution analysis is reduced, update iteration of research problems and experimental schemes on a time sequence is accurately described, and key research problems and core experimental schemes are found. Aiming at an alternate evolution mechanism between a scientific experiment exploration research problem and an experiment scheme, the limitation that evolution analysis focuses on single dimension, is insufficient in semantics, single in evolution dimension, one-sided in analysis and the like is overcome.

Description

Technical Field

[0001] The present application relates to the field of computer software technology, and more specifically, to a method and device for analyzing the evolution of scientific experiments based on a knowledge graph. Background Art

[0002] The new generation of artificial intelligence technology represented by large language models provides new technologies and methods for data-driven scientific research. Digital intelligence-driven scientific research and knowledge discovery have become new engines of scientific and technological innovation and are gradually becoming a new scientific research paradigm. Scientific experiments are a key link in scientific research, and research problems and experimental plans are the core content of scientific experiments. Evolutionary analysis is one of the typical knowledge discovery scenarios. Through the evolutionary analysis of scientific experiments, we can deeply explore the evolutionary development of scientific experiments, analyze the alternation and integration of problems and plans, and identify key research problems and core experimental plans in the field.

[0003] Commonly used technology evolution analysis methods today include methods based on bibliometrics, methods based on social network analysis, methods based on text mining, and personalized combinations of the above three methods. The method based on bibliometrics was first used in technology evolution analysis. It mainly uses common bibliometric methods such as co-occurrence analysis and co-citation analysis to construct technology evolution paths and analyze technology evolution processes. For example, Small et al. conducted knowledge diffusion research based on co-citation analysis. This type of method is simple in principle and easy to obtain results, but it does not go deep into the text content, the characterization of technology evolution is relatively rough, and it is not easy to directly mine the evolution path. Therefore, some researchers began to try to use social network analysis methods to mine the citation relationship between documents and directly characterize the evolution path. Common methods include main path analysis and network topology clustering. For example, Kim et al. characterized the evolution path of high-voltage direct current transmission technology based on main path analysis. The network topology clustering method can further mine the citation network and identify citation clusters. For example, Chen et al. used the GN community discovery algorithm to discover the citation network community, identify the technology clusters of fuel cell technology, and conduct evolution analysis. However, such methods still fail to directly mine text content. Therefore, researchers currently use more technology evolution analysis based on text mining, through keyword extraction or LDA topic modeling combined with temporal relationships. For example, Xu et al. constructed a linkage evolution path for scientific and technological themes based on technologies such as LDA topic modeling and bibliometric indicators. In addition, some researchers have gradually begun to try to combine the above three methods, usually by first performing cluster analysis and then analyzing the development of technology in each category. For example, based on clustering and network visualization technology, characteristic indicators are combined to construct a technology evolution path and analyze its development trend.

[0004] However, existing evolutionary analysis research often focuses on analyzing the formal evolutionary processes of inheritance, splitting, merging, birth and extinction in a single dimension, which cannot reflect the development context of alternating evolution of research problems and experimental schemes in scientific experiments and mutual promotion of scientific and technological innovation. Summary of the invention

[0005] The purpose of this application is to provide a scientific experiment evolution analysis method and device based on knowledge graphs in order to overcome the existing technical defects, realize the quantitative calculation of the evolution process, reduce the subjectivity of evolution analysis, accurately describe the update and iteration of research problems and experimental schemes in time series, and discover key research problems and core experimental schemes. It can also explore the alternating evolution mechanism between research problems and experimental schemes for scientific experiments, which can overcome the current evolution analysis that is mainly single-dimensional, has the limitations of insufficient semantics, single evolutionary dimension, and one-sided analysis.

[0006] The purpose of this application is achieved through the following technical solutions: In the first aspect, the present application proposes a scientific experiment evolution analysis method based on knowledge graph, the method comprising: Vectorize the scientific experiment according to the constructed scientific experiment knowledge graph to obtain the scientific experiment after vectorization; The community discovery algorithm is used to perform cluster analysis on the scientific experiments after vectorization to obtain the clustering results; Model the clustering results and construct a scientific experiment evolution map; The evolution results of scientific experiments are obtained by performing point and surface analysis on the evolution map of scientific experiments.

[0007] In one possible implementation, the scientific experiment knowledge graph is constructed by using knowledge graph technology to perform fine-grained, multi-dimensional knowledge representation and modeling of complex knowledge objects, temporal knowledge objects, and knowledge and data fusion knowledge objects in scientific experiments.

[0008] In a possible implementation, the steps of vectorizing the scientific experiment according to the constructed scientific experiment knowledge graph to obtain the vectorized scientific experiment include: Classify scientific experiments according to the constructed scientific experiment knowledge graph to obtain research questions and experimental plans; The research problem and experimental plan, the knowledge and data of the scientific experiment knowledge graph are respectively vectorized to obtain the scientific experiment after vectorization.

[0009] In one possible implementation, a long short-term memory network is used to vectorize the knowledge of the scientific experiment knowledge graph.

[0010] In a possible implementation, the data of the scientific experiment knowledge graph is vectorized using corresponding unit vectors.

[0011] In a possible implementation, the research problem includes a general research problem and a specific research problem. The vectorized representation of the general research problem is: ; in For general questions, The number of specific questions included in the general question, is the amount of knowledge contained in the specific problem, is the knowledge representation vector, is the data representation vector, All are weight coefficients; The vectorized representation of the specific research problem is: ; in Represents a specific issue, is the amount of knowledge contained in the specific problem, is the amount of data included in the specific question, is the knowledge representation vector for a specific research problem, is the data representation vector for a specific research question, For the general problem representation corresponding to the specific problem, and are all weight coefficients.

[0012] In a possible implementation, the experimental scheme includes a general experimental scheme and a specific experimental scheme, and the vectorized expression of the general experimental scheme is: The vectorized representation of the general experimental scheme = [[the weighted sum of all experimental target knowledge representations of the general experimental scheme, the weighted sum of all experimental principle knowledge representations of the general experimental scheme, the weighted sum of all experimental method knowledge representations of the general experimental scheme, the weighted sum of all experimental step knowledge representations of the general experimental scheme], the weighted sum of the data representations of all specific schemes of the general experimental scheme]; The vectorized representation of the specific experimental scheme is:

[0013] ,in Represents a specific experimental plan. Represents the experimental goal corresponding to the specific experimental plan, Represents the experimental method corresponding to the specific experimental plan, Represents the experimental steps corresponding to the specific experimental plan, represents the number of knowledge representations corresponding to a specific experimental scheme, , , , and They represent the corresponding knowledge vectors, Represents the number of data knowledge representations corresponding to the experimental scheme, are weight coefficients, It is the knowledge representation of the general experimental plan corresponding to the specific experimental plan.

[0014] In a possible implementation, the community discovery algorithm uses the Leiden algorithm, and the step of using the community discovery algorithm to perform cluster analysis on the vectorized scientific experiments to obtain clustering results includes: Calculate the cosine similarity between the research question and the experimental plan, and remove the cosine similarity below the similarity threshold to obtain the cosine similarity after the change; The cosine similarity after the change is used as the edge weight to construct an undirected weighted graph and input it into the Leiden algorithm; First cluster the research questions using the Leiden algorithm, and then select the experimental plans corresponding to the specific research questions contained in the key research questions; The Leiden algorithm is used to cluster the experimental schemes into core experimental schemes of corresponding categories to obtain clustering results.

[0015] In a possible implementation, the step of modeling the clustering results and constructing a scientific experiment evolution map includes: Construct evolutionary semantic relationships between research questions, between experimental schemes, and between research questions and experimental schemes based on the similarities in the clustering results; Draw a scientific experiment evolution map in chronological order based on evolutionary semantic relationships.

[0016] In a second aspect, the present application also proposes a scientific experiment evolution analysis device based on a knowledge graph, the device comprising: A vectorization module is used to vectorize the scientific experiment according to the constructed scientific experiment knowledge graph to obtain the scientific experiment after vectorization; The clustering analysis module is used to perform clustering analysis on the scientific experiments after vectorization representation using the community discovery algorithm to obtain clustering results; Modeling module, used to model clustering results and construct scientific experimental evolution maps; The result generation module is used to perform point and surface analysis on the scientific experiment evolution map to obtain the scientific experiment evolution results.

[0017] The above main scheme of the present application and its further options can be freely combined to form multiple schemes, all of which are schemes that can be adopted and claimed for protection in the present application; and in the present application, (non-conflicting options) options and other options can also be freely combined. After understanding the scheme of the present application, those skilled in the art can understand that there are many combinations based on the existing technology and common knowledge, all of which are technical schemes to be protected by the present application, and they are not exhaustively listed here.

[0018] The present application discloses a method and device for analyzing the evolution of scientific experiments based on a knowledge graph. First, the scientific experiment is vectorized according to the constructed scientific experiment knowledge graph to obtain the scientific experiment after vectorization. Secondly, a community discovery algorithm is used to perform cluster analysis on the scientific experiment after vectorization to obtain clustering results. The clustering results are modeled to construct a scientific experiment evolution graph. Finally, the scientific experiment evolution graph is analyzed in terms of points and surfaces to obtain the scientific experiment evolution results. Quantitative calculation of the evolutionary process is achieved, the subjectivity of evolutionary analysis is reduced, the update and iteration of research problems and experimental schemes in time series are accurately portrayed, and key research problems and core experimental schemes are discovered. It is also possible to explore the alternating evolution mechanism between research problems and experimental schemes for scientific experiments, which can overcome the limitations of current evolutionary analysis, which is mainly single-dimensional, with insufficient semantics, a single evolutionary dimension, and one-sided analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0020] Figure 1 A flow chart of the scientific experiment evolution analysis method based on knowledge graph proposed in an embodiment of the present application is shown.

[0021] Figure 2 A schematic diagram of a three-layer data fusion architecture proposed in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0022] The following describes the embodiments of the present application through specific examples, and those skilled in the art can easily understand other advantages and effects of the present application from the contents disclosed in this specification. The present application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present application. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0023] Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of this application.

[0024] Scientific experiment evolution analysis is a complex knowledge discovery scenario. In the existing technology, the research on evolution analysis has the following deficiencies: First, at the data level, scientific experiments include complex knowledge objects such as research questions and experimental plans. Experimental plans also include temporal knowledge objects such as experimental steps and scientific experimental processes, as well as knowledge and data fusion knowledge objects such as experimental elements. Traditional methods mostly use keywords to model scientific experiments, which have weak semantics and strong subjectivity, and it is difficult to accurately represent various comprehensive knowledge objects in scientific experiments.

[0025] Second, at the level of evolutionary analysis methods, it is usually necessary to integrate research problems with experimental plans for evolutionary analysis in order to explore the development context of alternating evolution of problems and plans and mutual promotion of scientific and technological innovation. Traditional evolutionary research mainly analyzes the formal inheritance, splitting, merger, birth and extinction of evolutionary subjects in a single dimension, which is not suitable for studying the development context of multi-dimensional alternating evolution of problems and plans in scientific experiments and mutual promotion of scientific and technological innovation.

[0026] Therefore, in order to solve the above problems, the embodiment of the present application proposes a scientific experiment evolution analysis method and device based on knowledge graph, which uses the current mainstream evolution analysis method based on text mining and network analysis combined technology, fully analyzes the evolution semantic relationship between research problems and experimental schemes in scientific experiments, such as update and evolve, and constructs a panoramic cross-evolution map of scientific experiments from the two dimensions of research problems and experimental schemes. Next, a scientific experiment evolution analysis method and device based on knowledge graph proposed in the embodiment of the present application are described in detail.

[0027] Please refer to Figure 1 , Figure 1 The flowchart of the scientific experiment evolution analysis method based on knowledge graph proposed in the embodiment of the present application is shown, which includes the following steps: Step S100: vectorize the scientific experiment according to the constructed scientific experiment knowledge graph to obtain the vectorized scientific experiment.

[0028] The scientific experiment knowledge graph is constructed in the following way: knowledge graph technology is used to perform fine-grained, multi-dimensional knowledge representation and modeling on complex knowledge objects, temporal knowledge objects, and knowledge and data fusion knowledge objects in scientific experiments.

[0029] Knowledge graph representation of scientific experiments: Use knowledge graph technology to perform fine-grained, multi-dimensional knowledge representation and modeling of complex knowledge objects, temporal knowledge objects, and knowledge and data fusion knowledge objects contained in scientific experiments, and construct a scientific experiment knowledge graph. Then, based on the idea of ​​hierarchical vectorization, combine the knowledge and data in the scientific experiment knowledge graph to vectorize the scientific experiment to provide rich semantics, provide a data basis for the quantitative calculation of evolutionary semantic relationships, solve the problem of strong dependence on manual labor in previous studies, and provide basic layer data for the identification of key research problems and core experimental plans.

[0030] Step S100 includes: Classify scientific experiments according to the constructed scientific experiment knowledge graph to obtain research questions and experimental plans; The research problem and experimental plan, the knowledge and data of the scientific experiment knowledge graph are respectively vectorized to obtain the scientific experiment after vectorization.

[0031] Based on the idea of ​​hierarchical vectorization, the knowledge and data in the scientific experiment knowledge graph are combined to vectorize the scientific experiments, making the evolution more comprehensive and multi-dimensional, and providing rich semantics to meet the objectivity of academic research. It can solve the problems of the existing technology of using keywords for knowledge representation, strong dependence on manual work, weak semantics and strong subjectivity.

[0032] In order to conduct scientific experimental evolution analysis, based on Figure 1 Please refer to Figure 2 , Figure 2 A schematic diagram of a three-layer data fusion architecture proposed in an embodiment of the present application is shown. Using the constructed scientific experiment knowledge graph, scientific experiments are divided into research problems and experimental schemes and vectorized to provide basic layer data. In order to efficiently perform scientific experiment evolution analysis, similar scientific experiments are clustered from the two dimensions of problems and schemes, and key research problems and core experimental schemes are preliminarily explored to provide analysis layer data. Fully analyze the semantic relationships such as update and evolve between research problems and experimental schemes, construct a scientific experiment evolution map to provide application layer data, and identify key research problems and core experimental schemes in the field. Finally, the results of the scientific experiment evolution are analyzed and verified, which can highlight the advantages of technologies such as event knowledge graphs, community discovery, and multi-dimensional cross-evolution in the analysis of scientific experiment evolution.

[0033] In terms of knowledge representation, existing research mainly uses the keyword method for knowledge representation, and it is highly dependent on manual work, with shortcomings such as weak semantics and strong subjectivity. The layered vectorization idea combines the knowledge and data in the scientific experiment knowledge graph to vectorize the scientific experiment, making the evolution more comprehensive and multi-dimensional, and providing rich semantics to meet the objectivity of academic research.

[0034] The purpose of the vectorized representation of scientific experiments is to provide a data basis for the quantitative calculation of evolutionary semantic relationships, solve the problem of strong dependence on artificial intelligence in previous studies, and provide basic layer data for the identification of key research problems and core experimental schemes. All static entities and dynamic event information directly related to scientific experiments in the scientific experiment knowledge graph will be used to vectorize scientific experiments, and the research problems and experimental schemes in scientific experiments will be vectorized, which can be divided into micro-level and meso-level representations.

[0035] The vectorization representation at the micro level includes the vectorization of knowledge and data in the scientific experiment knowledge graph, which is divided into SEKG knowledge vectorization and SEKG data vectorization. SEKG (Semantic Enhanced Knowledge Graph Construction) knowledge vectorization refers to the process of converting entities and relationships in the knowledge graph into numerical vector representations. SEKG data vectorization refers to the process of converting non-numerical data into numerical vector form so that machine learning models or deep learning networks can process these data.

[0036] In one possible implementation, a long short-term memory network is used to vectorize the knowledge of the scientific experiment knowledge graph.

[0037] In the vectorization of knowledge, based on the long sequence and time dynamics of scientific experimental knowledge, combined with the advantages of LSTM in processing input sequences, such as effective capture of long-term dependencies, gradient stability and wide applicability, the loss value of each batch is saved during the operation of the LSTM program, and the loss value change curve of each round of word vector training can be obtained. By analyzing the curve graph, the loss value change trend of the model can be obtained, reflecting the accuracy of word vector training. Use LSTM to vectorize scientific experimental knowledge. Table 1 shows an example of parameter settings for LSTM model parameters: Table 1

[0038] In a possible implementation, the data of the scientific experiment knowledge graph is vectorized using corresponding unit vectors.

[0039] In the vectorization of data, the unit vectors corresponding to high, medium and low are used to uniformly represent the scientific experimental data (high corresponds to [1,0,0], medium corresponds to [0,1,0], and low corresponds to [0,0,1]), which can effectively solve the PCE / / / FF indicator unit is not unified, Table 2 numerical data vectorization corresponding range: Table 2

[0040] Research questions include general research questions and specific research questions.

[0041] Vectorization of general research questions: In the event ontology of the scientific experiment knowledge graph, 8 types of research questions and 5 types of experimental plans are constructed. In order to fully semantically represent the research questions and experimental plans corresponding to each scientific experiment, we hereby model the vectorized representation of 8 types of general research questions and 5 types of general experimental plans at the meso level, laying the foundation for the subsequent vectorized representation of each specific research question and experimental plan.

[0042] A general research problem is represented by the weighted sum of the knowledge and data of all specific research problems corresponding to this type of problem. Specifically, the vectorized representation of a general problem = [the weighted sum of the knowledge representation of all specific problems corresponding to this type of problem, the weighted sum of the data representation of all specific problems corresponding to this type of problem]. The vectorized representation of a general research problem is: ; in For general questions, The number of specific questions included in the general question, is the amount of knowledge contained in the specific problem, is the knowledge representation vector, is the data representation vector, are all weight coefficients.

[0043] Assuming a specific problem ( ), then N is The amount of knowledge contained, .

[0044] Vectorization of general experimental protocols: The general experimental plan is represented by the weighted sum of the knowledge and data of all specific experimental plans corresponding to this type of experimental plan, where the knowledge part includes the experimental objectives, experimental principles, experimental methods and experimental steps involved in the experimental plan.

[0045] The vectorized representation of the general experimental scheme is: The vectorized representation of the general experimental scheme = [[the weighted sum of all experimental target knowledge representations of the general experimental scheme, the weighted sum of all experimental principle knowledge representations of the general experimental scheme, the weighted sum of all experimental method knowledge representations of the general experimental scheme, the weighted sum of all experimental step knowledge representations of the general experimental scheme], the weighted sum of the data representations of all specific schemes of the general experimental scheme]; The data of all specific solutions of this type are weighted summation, see the following formula: ,in , , , , , represents a general experimental scheme, Represents the weighted sum of all experimental objectives of this type of experiment. Represents the weighted sum of all experimental principles of this type of experiment, Represents the weighted sum of all experimental methods of this type of experiment, Represents the weighted sum of all experimental steps in this type of experiment. Represents the weighted sum of all data vectors of this type of experiment, are weight coefficients, .

[0046] R, T, U, and V represent the experimental objectives corresponding to a specific experimental plan (consisting of R Knowledge vector weighted representation), experimental principle (by T Knowledge vector weighted representation), experimental method (by U knowledge vector weighted representation) and experimental steps (represented by V The number of knowledge representations (weighted representation of knowledge vectors), W represents the data knowledge corresponding to the experimental scheme (consisting of W Knowledge vector weighted representation) represents quantity.

[0047] The vectorized representation of a research problem corresponding to a specific scientific experiment consists of three parts, namely knowledge, data, and general research problems. The vectorized representation of a specific research problem is: ; in Represents a specific issue, is the amount of knowledge contained in the specific problem, is the amount of data included in the specific question, is the knowledge representation vector for a specific research problem, is the data representation vector for a specific research question, For the general problem representation corresponding to the specific problem, and are weight coefficients, , .

[0048] The vectorized representation of the experimental plan corresponding to a specific scientific experiment also includes three parts: knowledge, data, and the corresponding general experimental plan. The vectorized representation of the specific experimental plan is:

[0049] ,in Represents a specific experimental plan. Represents the experimental goal corresponding to the specific experimental plan, Represents the experimental method corresponding to the specific experimental plan, Represents the experimental steps corresponding to the specific experimental plan, represents the number of knowledge representations corresponding to a specific experimental scheme, , , , and They represent the corresponding knowledge vectors, Represents the data knowledge corresponding to the experimental scheme (consisting of W Knowledge vector weighted representation) represents quantity, are weight coefficients, It is the knowledge representation of the general experimental plan corresponding to the specific experimental plan.

[0050] It is worth mentioning that the above specific research questions, experimental plans and vectorized representations are all custom data structures, each containing three parts, and are represented separately rather than being integrated into a whole.

[0051] Step S200: Use a community discovery algorithm to perform cluster analysis on the scientific experiments after vectorization to obtain clustering results.

[0052] Clustering of scientific experiments: A community discovery algorithm is used to perform cluster analysis on scientific experiments. In order to more effectively discover key research issues and core experimental schemes in scientific experiments, the Leiden algorithm, one of the current SOTA community discovery algorithms, is used to perform cluster analysis on scientific experiments. This algorithm has the advantages of high efficiency, robustness, and good interpretability. It can effectively meet the characteristics of scientific experiments and the conditions required for their clustering. The algorithm will automatically iterate until the modularity converges and output the best clustering effect. The key research issues and core experimental schemes in the field of organic solar cells are preliminarily obtained, providing analysis layer data for the final identification of key research issues and core experimental schemes, which can support evolutionary analysis.

[0053] Step S200 includes: Calculate the cosine similarity between the research question and the experimental plan, and remove the cosine similarity below the similarity threshold to obtain the cosine similarity after the change; The cosine similarity after the change is used as the edge weight to construct an undirected weighted graph and input it into the Leiden algorithm; First cluster the research questions using the Leiden algorithm, and then select the experimental plans corresponding to the specific research questions contained in the key research questions; The Leiden algorithm is used to cluster the experimental schemes into core experimental schemes of corresponding categories to obtain clustering results.

[0054] Undirected weighted graphs of research questions and experimental schemes are constructed respectively and input into the Leiden algorithm. The similarity between research questions and experimental schemes is calculated using cosine similarity, and similarity values ​​below the similarity threshold are eliminated. The similarity that satisfies the conditions is used as the edge weight between research questions and between experimental schemes. In this way, two undirected weighted graphs are constructed as inputs of the Leiden algorithm. The input examples are shown in Table 3: Table 3

[0055] While calculating the similarity, the centrality of each research question and experimental scheme is also calculated to determine the labels of each category obtained by clustering. Specifically, the node with the largest centrality value in each category is selected as the label of the category.

[0056] Next, we use the Leiden algorithm to cluster all the research questions and get the key research questions. , and then screen the experimental plans corresponding to the specific research questions included in all key research questions And cluster the n experimental schemes into core experimental schemes of corresponding categories using Leiden algorithm , and obtain the clustering result output by the Leiden algorithm.

[0057] The Leiden algorithm based on modularity optimization is used to identify key research problems and core experimental schemes of scientific experiments. The clustering results between problems, schemes, and problems reveal the main research trends and experimental techniques in the field, and combined with expert consultation, it can provide directions for future research in the discipline.

[0058] Step S300: Model the clustering results and construct a scientific experiment evolution map.

[0059] Step S300 includes: Construct evolutionary semantic relationships between research questions, between experimental schemes, and between research questions and experimental schemes based on the similarities in the clustering results; Draw a scientific experiment evolution map in chronological order based on evolutionary semantic relationships.

[0060] By calculating the cosine similarity between problems, solutions, and problems (clustering results), the calculation results of the evolutionary association between problems and solutions are obtained as shown in Table 4: Table 4

[0061] The calculation results of the above evolutionary associations are filtered to obtain the Same relationships between problems and between solutions. The specific introduction of the Same relationship is shown in Table 5, the introduction to the evolutionary semantic relationship.

[0062] Same association filtering rules are: set similarity threshold , if the maximum similarity Less than , then the association is deemed invalid. After filtering, there is still an evolutionary correlation with the remaining issues. With the current question only The similarity with other questions is greater than or equal to the evolution correlation threshold threshold, which is calculated as: .

[0063] Where N is the number of The number of issues associated with it, For some kind of problem With a specific problem The similarity between is the weight threshold, for The maximum similarity with the rest of the questions.

[0064] The above steps can filter out the Same relationships between problems and between solutions.

[0065] Then, the evolutionary semantic relationship is constructed. Since the traditional clustering results include birth, extinction, inheritance, splitting, merging, etc., such clustering results only construct evolutionary associations based on similarity and other methods, and the results are interpreted in form, but the specific semantic relationship between evolutionary objects is not established based on the evolutionary association in terms of content. Therefore, referring to the existing research results on technology evolution, the evolutionary semantic relationship between the research questions and experimental schemes used in Table 5 is given: Table 5

[0066] By using a self-written Python program, we can quantitatively calculate the Same / Update relationship between the research questions in Table 6, the Same / Update relationship between the experimental schemes in Table 7, and the Solve / Evolve relationship between the research questions and experimental schemes in Table 8: Table 6

[0067] Table 7

[0068] Table 8

[0069] The graph visualization presents the semantic relationship between the update, evolve and other event evolutions between the two types of events: research problems and experimental plans in scientific experiments. It can explore the development context of the alternating evolution of problems and plans and the mutual promotion of scientific and technological innovation, and provide application-layer data for the final identification of key research problems and core experimental plans in the field.

[0070] In addition, the Leiden algorithm removes outliers in the undirected weighted graph of research questions and experimental schemes during operation. Considering that these outliers are very likely to be emerging research questions or experimental schemes that suddenly appear in the subject field, these problems or schemes may also introduce research methods from other disciplines. Although these outliers have not attracted the attention of researchers in the short term, they may bring great breakthroughs to the development of the discipline. Therefore, experts in the subject field can be further invited to judge the outliers in the research questions and the outliers in the experimental schemes, and analyze emerging / interdisciplinary research questions and emerging / interdisciplinary experimental schemes.

[0071] In constructing clustering results, existing research focuses on analyzing the formal evolutionary process of inheritance, splitting, merging, birth and extinction in a single dimension, which cannot reflect the development context of alternating evolution of research problems and experimental schemes in scientific experiments and mutual promotion of scientific and technological innovation.

[0072] Construction of scientific experiment evolutionary map: Design evolutionary semantic relations such as evolve and update, and model the clustering results between problems and problems, solutions and solutions, and problems and solutions based on the current mainstream evolutionary analysis method based on a combination of text mining and network analysis. Construct a scientific experiment evolutionary map to provide application-layer data for the ultimate identification of key research issues and core experimental solutions in the discipline.

[0073] After calculating the cosine similarity between clustered problems, solutions, and problems, a similarity threshold is set, and after eliminating invalid associations, the corresponding evolution results are retained. Time is introduced to construct a scientific experiment evolution map and analyze the results, providing application-layer data for the final identification of key research issues and core experimental plans in the field.

[0074] The evolution of scientific experiments in the embodiments of this application can be summarized into technology evolution analysis, which can analyze the processes of generation, citation, iteration, extension and change of research problems and experimental schemes in scientific experiments, reveal the development and current status of scientific experiments in the form of evolutionary paths, and further identify and locate hot issues and key solutions.

[0075] In constructing the evolutionary map, experts are first asked to eliminate evolutionary semantic relationships that are obviously erroneous, retain evolutionary semantic relationship mining results that have certain research value, and draw a scientific experimental evolutionary map in chronological order.

[0076] Step S400: Analyze the scientific experiment evolution map in terms of points and surfaces to obtain the scientific experiment evolution results.

[0077] Analysis of the results of scientific experiment evolution: point and surface analysis are performed on the scientific experiment evolution map. The point analysis refers to the interpretation of the single-dimensional evolution results of the research problem, explaining the update and iteration process of such problems driven by the evolution of experimental schemes, and identifying the key research problems in the discipline field in combination with the overall evolution map; the surface analysis refers to the interpretation of the evolution results by combining such research problems with the panoramic cross-evolution map of the corresponding experimental scheme, explaining the development context of the multi-dimensional alternating evolution of research problems and experimental schemes and mutual promotion of scientific and technological innovation, and identifying the core experimental schemes in the discipline field in combination with the overall evolution map.

[0078] The results of evolutionary semantic mining are analyzed and verified for accuracy through expert verification. For example, the accuracy of evolutionary semantic relationships can be divided into 6 levels, namely A (accuracy 100%), B (accuracy 80%), C (accuracy 60%), D (accuracy 40%), E (accuracy 20%), and F (accuracy 0%). Each level can be defined as the result of the evaluation of the degree of evolution between Problem and Solution based on the clear clustering results. The higher the level, the higher the degree of evolution, and the lower the level, the lower the degree of evolution (for example, there is an Update relationship between P1 and P2, and the expert determines that the level is C, which means that about 60% of P1 evolves into P2, and the rest of P2 comes from the evolution of other Problems or Solutions).

[0079] In summary, the scientific experiment evolution analysis method proposed in the embodiment of the present application can quantitatively calculate the evolution process of scientific experiments, can mine the evolutionary semantic relationship between research problems and experimental schemes in scientific experiments, and construct a scientific experiment evolution map in multiple dimensions. It is highly efficient and has been verified by experts to accurately depict the update and iteration of research problems and experimental schemes in time series, reveal the development and current status of scientific experiments in the form of cross-evolutionary paths, and can further identify and locate key issues and core solutions. It overcomes the limitations of traditional evolutionary analysis, such as insufficient semantics, single evolutionary dimension, and one-sided analysis results, and meets new scenarios and new needs for knowledge discovery such as research on the evolutionary mechanism of scientific experiments, and has practical application value.

[0080] Compared with the prior art, the embodiments of the present application have the following beneficial effects: First, it realizes the quantitative calculation of the evolutionary process, reduces the subjectivity of evolutionary analysis to a certain extent, accurately describes the update and iteration of research problems and experimental schemes in time series, and discovers key research problems and core experimental schemes.

[0081] Second, it can explore the alternating evolutionary mechanism between research problems and experimental plans, serve scientific and technological innovation, and promote the development of the scientific research system. The study of evolutionary mechanisms can overcome the limitations of current evolutionary analysis, which is mainly single-dimensional, with insufficient semantics, single evolutionary dimension, and one-sided analysis.

[0082] Third, through an innovative three-layer data fusion architecture, the scientific experiment knowledge graph is deeply integrated into vectorized data, clustering data and evolution graph data, providing data support for the basic layer, analysis layer and application layer respectively. It has the advantages of traceable data, explainable process and easy reuse of results, providing strong support for digital intelligence-driven knowledge discovery scenario applications.

[0083] The following is a possible implementation of a scientific experiment evolution analysis device based on a knowledge graph, which is used to execute the various execution steps and corresponding technical effects of the scientific experiment evolution analysis method shown in the above embodiments and possible implementations. The device includes: A vectorization module is used to vectorize the scientific experiment according to the constructed scientific experiment knowledge graph to obtain the scientific experiment after vectorization; The clustering analysis module is used to perform clustering analysis on the scientific experiments after vectorization representation using the community discovery algorithm to obtain clustering results; Modeling module, used to model clustering results and construct scientific experimental evolution maps; The result generation module is used to perform point and surface analysis on the scientific experiment evolution map to obtain the scientific experiment evolution results.

[0084] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A scientific experiment evolution analysis method based on knowledge graph, characterized in that: The method comprises: Vectorize the scientific experiment according to the constructed scientific experiment knowledge graph to obtain the scientific experiment after vectorization; The community discovery algorithm is used to perform cluster analysis on the scientific experiments after vectorization to obtain the clustering results; Model the clustering results and construct a scientific experiment evolution map; The evolution results of scientific experiments are obtained by performing point and surface analysis on the evolution map of scientific experiments.

2. The scientific experiment evolution analysis method according to claim 1, characterized in that: The construction method of scientific experiment knowledge graph is: using knowledge graph technology to perform fine-grained, multi-dimensional knowledge representation and modeling of complex knowledge objects, temporal knowledge objects, and knowledge and data fusion knowledge objects in scientific experiments.

3. The scientific experiment evolution analysis method according to claim 1, characterized in that: The steps of vectorizing the scientific experiment according to the constructed scientific experiment knowledge graph to obtain the vectorized scientific experiment include: Classify scientific experiments according to the constructed scientific experiment knowledge graph to obtain research questions and experimental plans; The research problem and experimental plan, the knowledge and data of the scientific experiment knowledge graph are respectively vectorized to obtain the scientific experiment after vectorization.

4. The scientific experiment evolution analysis method according to claim 3, characterized in that: The long short-term memory network is used to vectorize the knowledge of the scientific experiment knowledge graph.

5. The scientific experiment evolution analysis method according to claim 3, characterized in that: The data of the scientific experiment knowledge graph is vectorized using corresponding unit vectors.

6. The scientific experiment evolution analysis method according to claim 3, characterized in that: The research questions include general research questions and specific research questions. The vectorized representation of general research questions is: where p i is a general problem, M is the number of specific problems contained in the general problem, N1 is the amount of knowledge contained in the specific problem, is the knowledge representation vector, is the data representation vector, a m , β n , σ m ,ω n All are weight coefficients; The vectorized representation of the specific research problem is: where p j represents a specific problem, N2 is the amount of knowledge contained in the specific problem, O is the amount of data contained in the specific problem, is the knowledge representation vector for a specific research problem, is the data representation vector for the specific research question, P k is the general problem representation corresponding to the specific problem, η n and θ o are all weight coefficients.

7. The scientific experiment evolution analysis method according to claim 3, characterized in that: The experimental scheme includes a general experimental scheme and a specific experimental scheme. The vectorized expression of the general experimental scheme is: The vectorized representation of the general experimental plan = [[the weighted sum of all experimental target knowledge representations of the general experimental plan, the weighted sum of all experimental principle knowledge representations of the general experimental plan, the weighted sum of all experimental method knowledge representations of the general experimental plan, the weighted sum of all experimental step knowledge representations of the general experimental plan], the weighted sum of the data representations of all specific plans of the general experimental plan]; the vectorized representation of the specific experimental plan is: ， where s i represents a specific experimental plan, R represents the experimental goal corresponding to the specific experimental plan, T represents the experimental method corresponding to the specific experimental plan, U represents the experimental steps corresponding to the specific experimental plan, and V represents the number of knowledge representations corresponding to the specific experimental plan. and They represent the corresponding knowledge vectors respectively, W represents the number of data knowledge representations corresponding to the experimental scheme, α , r , , β , t , , σ , u , ,ω , v , , ,, w are weight coefficients, S k It is the knowledge representation of the general experimental plan corresponding to the specific experimental plan.

8. The scientific experiment evolution analysis method according to claim 1, characterized in that: The community discovery algorithm uses the Leiden algorithm. The steps of using the community discovery algorithm to perform cluster analysis on the scientific experiments after vectorization representation to obtain clustering results include: Calculate the cosine similarity between the research question and the experimental plan, and remove the cosine similarity below the similarity threshold to obtain the cosine similarity after the change; The cosine similarity after the change is used as the edge weight to construct an undirected weighted graph and input it into the Leiden algorithm; First cluster the research questions using the Leiden algorithm, and then select the experimental plans corresponding to the specific research questions contained in the key research questions; The Leiden algorithm is used to cluster the experimental schemes into core experimental schemes of corresponding categories to obtain clustering results.

9. The scientific experiment evolution analysis method according to claim 1, characterized in that: The steps of modeling the clustering results and constructing the scientific experiment evolution map include: Construct evolutionary semantic relationships between research questions, between experimental schemes, and between research questions and experimental schemes based on the similarities in the clustering results; Draw a scientific experiment evolution map in chronological order based on evolutionary semantic relationships.

10. A scientific experiment evolution analysis device based on knowledge graph, characterized in that: The device comprises: A vectorization module is used to vectorize the scientific experiment according to the constructed scientific experiment knowledge graph to obtain the scientific experiment after vectorization; The clustering analysis module is used to perform clustering analysis on the scientific experiments after vectorization representation using the community discovery algorithm to obtain clustering results; Modeling module, used to model clustering results and construct scientific experimental evolution maps; The result generation module is used to perform point and surface analysis on the scientific experiment evolution map to obtain the scientific experiment evolution results.

Images