Scientific research training resource data matching method and system based on knowledge graph

Abstract

The invention provides a scientific research and training resource data matching method and system based on a knowledge graph, and relates to the technical field of knowledge graphs. Feature extraction is performed on scientific research personnel data and training resource data by constructing a scientific research knowledge association structure comprising entity nodes, association edges and attribute tags; generating a personnel feature set and a resource feature set, performing cross-dimension matching calculation by using the association edge weight of the scientific research knowledge association structure, generating a matching association degree matrix, and generating a scientific research training resource matching result according to the matching association degree matrix and the dynamic updating rule of the scientific research knowledge association structure. Therefore, accurate description and cross-dimension matching of scientific research personnel and training resource characteristics are realized, demand changes and resource updating can be responded in real time, personalized and accurate training resource recommendation is provided for the scientific research personnel, and the scientific research training effect and efficiency are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of knowledge graph technology, and in particular to a method and system for matching scientific research and training resource data based on knowledge graphs. Background Art

[0002] In the field of scientific research, the effective matching of scientific research and training resources is crucial to improving the professional capabilities and scientific research efficiency of scientific researchers. However, traditional scientific research and training resource matching methods often rely on manual screening and empirical judgment. This method is not only inefficient, but also difficult to fully and accurately grasp the actual needs of scientific researchers and the characteristics of training resources. With the continuous accumulation of scientific research data and the rapid development of knowledge graph technology, how to use knowledge graphs to construct a scientific research knowledge association structure and realize intelligent matching of scientific research and training resources based on this structure has become an urgent problem to be solved. Although existing resource matching methods have attempted to introduce knowledge graph technology, most of them only stay at the simple information retrieval and recommendation level, lacking in-depth exploration of the characteristics of scientific researchers and training resources and cross-dimensional matching calculations, resulting in inaccurate matching results and unable to meet the diverse training needs of scientific researchers. Summary of the Invention

[0003] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a scientific research and training resource data matching method based on a knowledge graph, the method comprising: Constructing a scientific research knowledge association structure, wherein the scientific research knowledge association structure includes entity nodes, association edges, and attribute labels, wherein the entity nodes include personnel nodes, resource nodes, and knowledge nodes; Based on the scientific research knowledge association structure, feature extraction processing is performed on the scientific research personnel data to generate a personnel feature set, wherein the personnel feature set includes academic background features, learning trajectory features, and demand preference features; Based on the scientific research knowledge association structure, feature extraction processing is performed on the training resource data to generate a resource feature set, wherein the resource feature set includes content coverage features, difficulty gradient features, and adapted population features; Performing cross-dimensional matching calculation on the personnel feature set and the resource feature set through the associated edge weights of the scientific research knowledge association structure to generate a matching association matrix; According to the matching correlation matrix and the dynamic update rules of the scientific research knowledge association structure, a scientific research training resource matching result is generated.

[0004] On the other hand, an embodiment of the present invention also provides a scientific research and training resource data matching system based on a knowledge graph, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0005] Based on the above aspects, the embodiment of the present invention provides a comprehensive knowledge framework for feature extraction and matching of scientific researchers and training resources by constructing a scientific research knowledge association structure including entity nodes, associated edges and attribute labels. Based on the scientific research knowledge association structure, feature extraction processing is performed on scientific research personnel data and training resource data to generate a personnel feature set and a resource feature set containing features such as academic background, learning trajectory, demand preference, content coverage, difficulty gradient and adapted population, thereby achieving an accurate characterization of the characteristics of scientific researchers and training resources. Through the associated edge weights of the scientific research knowledge association structure, a cross-dimensional matching calculation is performed on the personnel feature set and the resource feature set to generate a matching correlation matrix, which fully considers the complex association relationship between scientific researchers and training resources and improves the accuracy and effectiveness of the matching. Finally, according to the dynamic update rules of the matching correlation matrix and the scientific research knowledge association structure, a scientific research training resource matching result is generated, which can respond to changes in scientific research personnel needs and updates of training resources in real time, provide scientific researchers with personalized and accurate training resource recommendations, and effectively improve the effect and efficiency of scientific research training. BRIEF DESCRIPTION OF THE DRAWINGS

[0006] Figure 1 It is a schematic diagram of the execution flow of the scientific research and training resource data matching method based on the knowledge graph provided in an embodiment of the present invention.

[0007] Figure 2 It is a schematic diagram of exemplary hardware and software components of a knowledge graph-based scientific research and training resource data matching system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0008] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 This is a flow chart of a method for matching scientific research and training resource data based on a knowledge graph provided by an embodiment of the present invention. The method for matching scientific research and training resource data based on a knowledge graph is introduced in detail below.

[0009] Step S110: constructing a scientific research knowledge association structure, wherein the scientific research knowledge association structure includes entity nodes, association edges and attribute labels, and the entity nodes include personnel nodes, resource nodes and knowledge nodes.

[0010] This embodiment aims to construct a scientific research knowledge association structure, which serves as the foundational framework for the entire scientific research and training resource data matching method. By defining entity nodes, associated edges, and attribute labels, it can present the complex relationships between people, resources, and knowledge in the scientific research field. Specifically, person nodes represent individuals participating in scientific research activities, resource nodes correspond to scientific research resources such as various training courses, and knowledge nodes cover various knowledge points within a discipline. These entity nodes are interconnected by associated edges, and attribute labels further describe the characteristics of the nodes and edges.

[0011] Step S111: Collect multi-source data from academic paper databases, training course platforms, and scientific research personnel management systems as a raw data set.

[0012] In order to build an accurate and comprehensive scientific research knowledge association structure, it is necessary to collect data from multiple data sources. The academic paper database is a common repository for scientific research results, which contains the latest research findings and academic opinions in various disciplines, and can reflect the cutting-edge dynamics and knowledge system of the discipline. The training course platform records a wealth of training resources, including the course content outline, teaching methods, applicable population, etc., which helps to understand the current status and needs of scientific research training. The scientific researcher management system stores the basic information, educational background, scientific research experience, and training records of scientific researchers, which helps to gain a deeper understanding of the academic background and development trajectory of scientific researchers.

[0013] Integrating the data from these three data sources creates a raw data set, which can provide rich material for subsequent entity recognition and relationship extraction. For example, papers in an academic paper database may involve interdisciplinary research, while courses on a training platform may target researchers at different levels. The data in the researcher management system can reflect the learning and research progress of researchers at different stages.

[0014] Step S112: Perform entity recognition processing on the original data set to extract personnel entities, resource entities and knowledge entities. The personnel entities correspond to individual scientific researchers, the resource entities correspond to specific training courses, and the knowledge entities correspond to subject knowledge points.

[0015] After obtaining the original data set, it needs to be processed for entity recognition in order to accurately identify personnel entities, resource entities, and knowledge entities from a large amount of data. For personnel entities, by analyzing the information in the scientific researcher management system, such as name, position, and affiliated institution, it is mapped to a specific individual scientific researcher. Resource entities are mainly extracted from the data of the training course platform, and the specific training course is determined based on information such as the course name, number, and content summary. The extraction of knowledge entities is relatively complex and requires in-depth analysis of the text content in the academic paper database and training course platform to identify the subject knowledge points involved, such as theoretical concepts, research methods, and technical means. For example, in an academic paper, a specific research method may be mentioned and identified as a knowledge entity; in the outline of a training course, the various knowledge points listed will also be extracted as knowledge entities.

[0016] Step S113: performing relationship extraction processing on the original data set to establish inter-entity association edges, wherein the association edges include the mastery relationship between the personnel entity and the knowledge entity, the coverage relationship between the resource entity and the knowledge entity, and the historical learning relationship between the personnel entity and the resource entity.

[0017] After entity recognition, the next step is relationship extraction to establish edges between entities. The mastery relationship between the person entity and the knowledge entity reflects a researcher's familiarity with and mastery of different knowledge points. By analyzing information such as a researcher's academic publications, research projects participated in, and training courses completed, their mastery of each knowledge entity can be inferred. For example, if a researcher applies a specific research method in multiple papers, they can be considered to have a good grasp of that knowledge entity. The coverage relationship between the resource entity and the knowledge entity reflects the scope of knowledge points covered in the training course. By analyzing the course syllabus, the knowledge entities covered in the course are determined, and thus the coverage relationship between the resource entity and the knowledge entity is established. The historical learning relationship between the person entity and the resource entity records the training courses that the researcher has participated in in the past. Through the training records in the researcher management system, it is possible to identify which researchers have taken which training courses, thereby establishing this historical learning relationship.

[0018] Step S114: performing attribute annotation processing on the entity node and setting attribute labels, wherein the attribute labels include the research direction of the personnel entity, the teaching form of the resource entity, and the knowledge level of the knowledge entity.

[0019] To describe the characteristics of entity nodes in more detail, attribute labeling is required. For person entities, research direction is a key attribute tag that reflects the researcher's primary research areas and interests. Research direction can be determined by analyzing information such as the topics of their academic papers, the research projects they participate in, and their positions within research institutions. The teaching format attribute tag for resource entities can be categorized as online, offline, or hybrid, helping researchers choose appropriate training courses based on their needs. The knowledge level attribute tag for knowledge entities is used to distinguish the difficulty and complexity of knowledge points, such as basic, intermediate, and advanced. The knowledge level of a knowledge entity can be determined by comprehensively considering factors such as its position within the disciplinary system, the depth of related research, and the breadth of its application.

[0020] Step S115: performing weight calibration processing on the associated edges, and calculating the associated edge weight values ​​based on the co-occurrence frequency, wherein the weight values ​​represent the closeness of the association between entities.

[0021] The weight of the associated edge is crucial for accurately reflecting the degree of association between entities. This step calculates the weight of the associated edge based on co-occurrence frequency. Co-occurrence frequency refers to the number of times two entities appear together in the original data. By counting co-occurrence frequency, we can quantify the closeness of the association between entities.

[0022] Step S1151: Count the number of co-occurrences of the person entity and the knowledge entity in the academic paper as the original weight value of the mastery relationship.

[0023] In academic papers, the co-occurrence of person entities and knowledge entities can reflect researchers' attention to and application of specific knowledge. By performing text analysis on papers in an academic paper database, we count the number of times each person entity and each knowledge entity co-occur. For example, if a researcher repeatedly mentions the knowledge entity "deep learning algorithm" in multiple academic papers on artificial intelligence, the co-occurrence count between this person entity and the knowledge entity "deep learning algorithm" will be relatively high. This co-occurrence count serves as the initial weight for the mastery relationship between them.

[0024] Step S1152: Count the number of co-occurrences of resource entities and knowledge entities in the course outline as the original weight value of the coverage relationship.

[0025] A course syllabus is a detailed plan of training course content, including the knowledge entities covered by the course. By analyzing the course syllabi on the training course platform, we count the number of times each resource entity (i.e., a specific training course) co-occurs with each knowledge entity. For example, in a training course syllabus on big data analytics, the knowledge entity "data mining algorithm" is mentioned multiple times. The number of co-occurrences between this resource entity and the knowledge entity "data mining algorithm" serves as the initial weight for the coverage relationship between them.

[0026] Step S1153: Count the number of co-occurrences of the personnel entity and the resource entity in the training records as the original weight value of the historical learning relationship.

[0027] The training records in the researcher management system record researchers' participation in training courses. By counting these records, we calculate the number of times each person entity co-occurs with each resource entity. For example, if a researcher participated in multiple machine learning training courses, the number of co-occurrences between the person entity and the machine learning training course resource entities would serve as the raw weight of the historical learning relationship between them.

[0028] Step S1154: normalize the original weight value to generate a standardized weight value in the range of 0-1.

[0029] Since the original weight values ​​of different associated edges may have different value ranges, in order to facilitate subsequent calculations and comparisons, the original weight values ​​need to be normalized. The purpose of normalization is to map the original weight values ​​to the range of 0-1. The specific method can adopt common normalization algorithms, such as Min-Max normalization. For the original weight value of each associated edge, subtract the minimum value in its set, and then divide it by the difference between the maximum and minimum values ​​in the set to obtain the standardized weight value. After this processing, the weight values ​​of all associated edges are between 0-1 and are comparable.

[0030] Step S1155: Smoothing the normalized weight value to generate a final associated edge weight value.

[0031] To prevent outliers in the normalized weights from significantly impacting subsequent calculations, they need to be smoothed. This can be accomplished using various smoothing algorithms, such as Gaussian smoothing or moving average smoothing. For example, for each normalized weight, the average of its adjacent weights is taken as the smoothed weight. This smoothing process makes the weights of associated edges more stable and reasonable, resulting in a final edge weight that more accurately reflects the closeness of the association between entities.

[0032] Step S120: Based on the scientific research knowledge association structure, feature extraction processing is performed on the scientific research personnel data to generate a personnel feature set, which includes academic background features, learning trajectory features and demand preference features.

[0033] After constructing the scientific research knowledge association structure, it is necessary to extract features from the researcher data to fully understand their characteristics and needs. The academic background feature in the personnel feature set can reflect the researcher's educational background and knowledge reserves. The learning trajectory feature can demonstrate the researcher's learning process and development trends. The demand preference feature reflects the researcher's expectations and needs for training resources.

[0034] Step S121: extracting educational background information, completed training records and training application text from the scientific researcher data as input data.

[0035] Researcher data contains a wealth of information, with educational background information, completed training records, and training application documents serving as key inputs for feature extraction. Educational background information includes the researcher's academic credentials, university of graduation, and major studied, reflecting their foundational knowledge and professional development. Completed training records document past training courses attended by researchers, including course titles, training times, and training outcomes. Analyzing these records allows us to understand their learning experiences and knowledge development. Training application documents are crucial for researchers to express their training needs and may include information such as the areas of knowledge they wish to learn and their desired training methods.

[0036] Step S122: Based on the mastery relationship between the personnel entity and the knowledge entity in the scientific research knowledge association structure, the educational background information is subjected to knowledge coverage analysis and processing, and the hierarchical distribution and quantitative proportion of the knowledge entities that the scientific researchers have mastered are extracted as academic background features.

[0037] By leveraging the mastery relationship between personnel entities and knowledge entities within the scientific research knowledge association structure, we conduct an in-depth analysis of educational background information. First, based on the researcher's academic credentials and major, we determine the range of knowledge entities they are likely to have access to and master. Then, combined with the knowledge hierarchy attributes of the knowledge entities, we analyze the distribution of knowledge entities mastered by researchers at different levels, as well as the percentage of knowledge entities at each level. For example, for a researcher with a master's degree in computer science, their mastery and percentage of knowledge entities at different levels, such as basic computer knowledge, intermediate algorithm design, and advanced artificial intelligence knowledge, constitute their academic background characteristics.

[0038] Step S123: Based on the historical learning relationship between the personnel entity and the resource entity in the scientific research knowledge association structure, a time series analysis is performed on the completed training records to extract the difficulty change trend of the scientific research personnel's learning resources and the knowledge coverage expansion path as learning trajectory features.

[0039] Time series analysis of completed training records can reveal the learning and development process of scientific researchers.

[0040] Step S1231: Arrange the completed training records of scientific researchers in chronological order to form a learning time series.

[0041] Arrange the researchers' completed training records in chronological order to form a learning time series. This learning time series can show the training courses that researchers have participated in at different time points.

[0042] Step S1232: For each training resource in the learning time series, extract the knowledge level span value of the difficulty gradient feature corresponding to it in the scientific research knowledge association structure.

[0043] For each training resource in the learning time series, we extract the knowledge level span value of the corresponding difficulty gradient feature based on the attribute information of the resource entity in the scientific research knowledge association structure. The knowledge level span value reflects the range of difficulty of the knowledge covered by the training course, from basic to advanced. For example, a training course that covers basic programming knowledge to advanced algorithm design has a relatively large knowledge level span value.

[0044] Step S1233: Calculate the difference in knowledge level span values ​​of adjacent training resources to generate a difficulty change rate sequence.

[0045] By calculating the difference in the knowledge span values ​​of two adjacent training resources in the learning time series, we can generate a difficulty change rate sequence. This difficulty change rate sequence can reflect the speed and trend of difficulty increase during the researcher's learning process. If the difference in the knowledge span values ​​of adjacent training resources is large, it means that the researcher will face a significant increase in difficulty in a short period of time; otherwise, it means that the difficulty increase is relatively gradual.

[0046] Step S1234: performing sliding window averaging processing on the difficulty change rate sequence to generate a difficulty change trend curve.

[0047] To more smoothly display the difficulty trend, we perform sliding window averaging on the difficulty change rate series. We select an appropriate window size and average the difficulty change rate values ​​within the window to obtain the average difficulty change rate corresponding to each window center point. These average difficulty change rates are then connected to form a difficulty change trend curve. This difficulty change trend curve can intuitively demonstrate the changing difficulty trend of researchers' learning resources and help determine the stability and development direction of their learning.

[0048] Step S1235: For each training resource in the learning time series, extract the knowledge association network corresponding to the content coverage feature in the scientific research knowledge association structure.

[0049] In addition to difficulty trends, we also need to analyze the knowledge expansion paths of researchers. For each training resource in the learning time series, we extract a knowledge association network based on its corresponding content coverage characteristics from the scientific research knowledge association structure. The knowledge association network displays the relationships between the knowledge entities involved in the training course, including the order of knowledge and causal relationships.

[0050] Step S1236: Through the node expansion analysis of the knowledge association network, the connection path of the knowledge coverage from the initial node to the newly added node in the learning process of the scientific researcher is extracted to generate the knowledge coverage expansion path.

[0051] We conduct node expansion analysis on the knowledge association network of each training resource, starting with the knowledge entities initially mastered by the researcher (i.e., initial nodes). We then gradually analyze the newly added knowledge entities (i.e., newly added nodes) during the learning process and identify the connection paths between them. These connection paths constitute the researcher's knowledge coverage expansion path, reflecting the accumulation and expansion of knowledge during the learning process.

[0052] Step S124: Based on the covering relationship between the knowledge entities and resource entities in the scientific research knowledge association structure, the training application text is semantically parsed to extract the knowledge entities to be learned and the expected resource forms mentioned by the researchers as demand preference features.

[0053] The training application text is semantically parsed, leveraging the overlapping relationships between knowledge entities and resource entities in the scientific research knowledge association structure to accurately extract the knowledge entities to be learned and the resource forms desired by the researchers. First, natural language processing is performed on the training application text to identify keywords and semantic information. These keywords are then matched with knowledge entities in the scientific research knowledge association structure to determine the knowledge areas that the researchers wish to learn. Furthermore, the text descriptions of training methods, course types, and other aspects are analyzed to extract the resource forms desired by the researchers, such as online video courses and offline lectures.

[0054] Step S125: combining the academic background features, the learning trajectory features, and the demand preference features into the personnel feature set.

[0055] The extracted academic background features, learning trajectory features, and demand preference features are combined to form a personnel feature set that comprehensively describes the characteristics and needs of researchers.

[0056] Step S130: Based on the scientific research knowledge association structure, feature extraction processing is performed on the training resource data to generate a resource feature set, which includes content coverage features, difficulty gradient features, and adapted population features.

[0057] To accurately match scientific research training resources with researchers, it's also necessary to extract features from the training resource data. The content coverage feature in the resource feature set reflects the scope of knowledge covered by the training resource, the difficulty gradient feature demonstrates the difficulty level and variation of the training resource, and the suitable population feature indicates the type of researcher the training resource is suitable for.

[0058] Step S131: extracting course outline text, historical student data and course evaluation records from the training resource data as input data.

[0059] Training resource data includes multiple aspects of information, including course syllabus text, historical student data, and course evaluation records, which are important inputs for feature extraction. The course syllabus text details the content and structure of the training course and is a key basis for determining content coverage characteristics. Historical student data records information about researchers who have participated in the training course, including their academic background and learning outcomes. Analyzing this data can help understand the demographics of the training resource. Course evaluation records reflect student satisfaction and feedback on the training course, helping to assess the quality and applicability of the training resource.

[0060] Step S132: Based on the coverage relationship between the resource entities and the knowledge entities in the scientific research knowledge association structure, the course outline text is subjected to knowledge node mapping processing, and the hierarchical distribution and association network of the course coverage knowledge entities are extracted as content coverage features.

[0061] The course outline text is mapped to knowledge nodes using the coverage relationship between resource entities and knowledge entities in the scientific research knowledge association structure. The knowledge points in the course outline text are matched with the knowledge entities in the scientific research knowledge association structure to determine the scope of knowledge entities covered by the course. Then, based on the knowledge hierarchy attributes of the knowledge entities, the distribution of knowledge entities covered by the course at different levels is statistically analyzed. At the same time, the association relationships between these knowledge entities are analyzed to construct a knowledge association network. For example, in a training course outline on bioinformatics, knowledge entities such as gene sequencing technology and biological data analysis algorithms are involved. Through mapping processing, their distribution at different levels and their mutual association relationships can be determined. The above information constitutes the content coverage characteristics of the course.

[0062] Step S133: Based on the knowledge level attributes of the knowledge entities in the scientific research knowledge association structure, the order of knowledge nodes in the course outline text is analyzed and processed in increasing difficulty, and the knowledge level span from basic to advanced and the connection strength of adjacent levels of the course are extracted as difficulty gradient features.

[0063] The order of knowledge nodes in the course syllabus text is analyzed in terms of increasing difficulty to determine the difficulty gradient characteristics of the course.

[0064] Step S1331: Identify the knowledge entities involved in the course outline text and obtain their knowledge hierarchical attribute values ​​in the scientific research knowledge association structure.

[0065] First, the course syllabus text is analyzed to identify the knowledge entities involved. Then, the knowledge-level attribute values ​​of these knowledge entities are obtained from the scientific research knowledge association structure. For example, a mathematics training course syllabus involves knowledge entities such as algebra, geometry, and calculus. By querying the scientific research knowledge association structure, the knowledge-level attribute values ​​of each knowledge entity can be obtained.

[0066] Step S1332: Arrange the knowledge entities in the order of the chapters in the course syllabus to form a knowledge node sequence.

[0067] The identified knowledge entities are arranged according to the chapter order of the course syllabus to form a knowledge node sequence, which can reflect the organization and logical order of the course content.

[0068] Step S1333: Calculate the difference between the knowledge level attribute values ​​of the first node and the last node in the knowledge node sequence as the knowledge level span.

[0069] The knowledge level span of a course is calculated by calculating the difference between the knowledge level attribute values ​​of the first and last nodes in a knowledge node sequence. This knowledge level span reflects the increasing difficulty of the course from basic to advanced. For example, a course that progresses from basic programming knowledge to advanced algorithm design has a relatively large knowledge level span.

[0070] Step S1334: performing difference calculation on the knowledge level attribute values ​​of adjacent nodes in the knowledge node sequence to generate a level-increasing difference sequence.

[0071] By calculating the difference between the knowledge level attribute values ​​of adjacent nodes in the knowledge node sequence, we can understand the increasing difficulty of the course content. Specifically, starting from the first node in the knowledge node sequence, calculate the difference between the knowledge level attribute values ​​of two adjacent nodes in sequence. For example, if the knowledge node sequence is node A, node B, node C, etc., calculate the knowledge level attribute value of node B minus the knowledge level attribute value of node A to get a difference; then calculate the knowledge level attribute value of node C minus the knowledge level attribute value of node B to get another difference, and so on. Arrange these differences in sequence to generate a hierarchical increasing difference sequence. Each difference in this hierarchical increasing difference sequence represents the degree of difficulty increase of the course between adjacent knowledge nodes.

[0072] Step S1335: Count the proportion of the hierarchical increasing difference sequence that is less than or equal to a preset threshold value as the connection strength of adjacent hierarchies.

[0073] The preset threshold is a reference standard designed based on the difficulty of the course. It is used to measure whether the connection between adjacent knowledge levels is smooth. Each difference in the sequence of increasing level differences is compared with the preset threshold, and the number of differences that are less than or equal to the preset threshold is counted. Then, the number of differences less than or equal to the preset threshold is divided by the total number of sequences of increasing level differences. The resulting ratio is the strength of the connection between adjacent levels. If this ratio is high, it means that the difficulty between adjacent knowledge levels of the course increases relatively gradually, and the connection is close; conversely, if the ratio is low, it means that the difficulty jumps between some adjacent knowledge levels of the course is large, and the connection is not smooth enough.

[0074] Step S1336: combining the knowledge level span and the connection strength into a difficulty gradient feature.

[0075] Combining the calculated knowledge level span and the strength of connections between adjacent levels forms the course's difficulty gradient. The knowledge level span reflects the overall difficulty progression from basic to advanced, while the strength of connections reflects the closeness of connections between adjacent knowledge levels within the course. Together, these two aspects constitute the difficulty gradient, comprehensively describing the course's difficulty and providing an important reference for researchers in selecting appropriate training courses.

[0076] Step S134: Based on the historical learning relationship between the personnel entity and the resource entity in the scientific research knowledge association structure, statistical analysis is performed on the academic background information in the historical student data, and the knowledge coverage level distribution and learning trajectory characteristics of the course history students are extracted as the adaptation population characteristics.

[0077] Leveraging the historical learning relationships between personnel entities and resource entities within the scientific research knowledge association structure, we conduct in-depth research on the academic background information within the history student data. First, based on information such as the history student's educational background and completed training courses, we determine the scope of knowledge entities they possess. Then, combining the knowledge hierarchy attributes of the knowledge entities, we statistically analyze the knowledge coverage of history students at different knowledge levels to obtain a knowledge coverage hierarchy distribution. Furthermore, we analyze the learning trajectories of history students, including the order in which they attended training courses and changes in difficulty, to extract learning trajectory features. These knowledge coverage hierarchy distributions and learning trajectory features together constitute the characteristics of the appropriate population for the course, which can help determine the type of scientific researchers for whom the course is suitable.

[0078] Step S135: combining the content coverage feature, the difficulty gradient feature, and the adapted population feature into the resource feature set.

[0079] The extracted content coverage features, difficulty gradient features, and applicable population features are combined to form a resource feature set, which comprehensively describes the characteristics and applicability of the training resources.

[0080] Step S140: performing cross-dimensional matching calculation on the personnel feature set and the resource feature set through the associated edge weights of the scientific research knowledge association structure to generate a matching association matrix.

[0081] After obtaining the personnel feature set and resource feature set, it is necessary to perform cross-dimensional matching calculations on these two sets using the associated edge weights of the scientific research knowledge association structure to determine the matching correlation between each scientific researcher and each training resource.

[0082] Step S141: extract the knowledge coverage level distribution of the academic background features in the personnel feature set, perform intersection calculation with the knowledge coverage level distribution of the content coverage features in the resource feature set, and generate a knowledge overlap index.

[0083] The knowledge coverage hierarchical distribution of academic background features is extracted from the personnel feature set. This distribution shows the knowledge mastery of researchers at different knowledge levels. Simultaneously, the knowledge coverage hierarchical distribution of content coverage features is extracted from the resource feature set, reflecting the content coverage of training resources at different knowledge levels. The intersection of these two knowledge coverage hierarchical distributions is calculated to identify the range of knowledge entities they jointly cover at the same knowledge level. A knowledge overlap index is generated by counting the number of jointly covered knowledge entities and weighting them with the associated edge weights. This knowledge overlap index reflects the degree of match between the researcher's academic background and the content of the training resources.

[0084] Step S142: extracting the difficulty change trend of the learning trajectory feature in the personnel feature set, performing trend consistency analysis on the knowledge level span of the difficulty gradient feature in the resource feature set, and generating a difficulty cohesion index.

[0085] The difficulty trend of the learning trajectory feature in the personnel feature set describes the increase in difficulty for researchers during their learning process, while the knowledge level span of the difficulty gradient feature in the resource feature set reflects the degree of difficulty increase from basic to advanced levels of the training resources. A trend consistency analysis was conducted on these two features to compare whether the difficulty trend of the researchers matches the knowledge level span of the training resources. For example, if the researcher's learning trajectory shows a relatively flat increase in difficulty, while the training resources have a larger knowledge level span, then their trend consistency may be low; conversely, if the trends are similar, then their trend consistency is high. Through a series of analyses and weighted calculations, a difficulty cohesion index was generated, which reflects the degree of match between the changes in the researchers' learning difficulty and the difficulty of the training resources.

[0086] Step S143: extracting the knowledge entities to be learned with the demand preference characteristics in the personnel feature set, performing path matching processing with the knowledge association network with the content coverage characteristics in the resource feature set, and generating a demand satisfaction index.

[0087] The researchers then extracted knowledge entities to be learned based on their needs and preferences from the personnel feature set. These knowledge entities represent the areas of knowledge that researchers wish to learn. Simultaneously, they extracted a knowledge association network based on content coverage from the resource feature set, which displays the relationships between the knowledge entities covered by the training resources. A path matching process was then performed to determine whether the training resource's knowledge association network contains the knowledge entities that researchers wish to learn and whether the paths connecting these knowledge entities within the network are reasonable. By counting the number of matched knowledge entities and the rationality of the paths, and combining this with the weights of the associated edges for weighted calculation, a demand satisfaction index was generated. This demand satisfaction index reflects whether the training resources meet the researchers' learning needs.

[0088] Step S144: extract the knowledge coverage ratio of the academic background features in the personnel feature set, calculate the distribution similarity with the knowledge coverage level distribution of the adapted population features in the resource feature set, and generate a population adaptation index.

[0089] The knowledge coverage ratio of the academic background feature in the personnel feature set reflects the proportion of knowledge mastered by researchers at different knowledge levels, while the knowledge coverage level distribution of the target population feature in the resource feature set shows the knowledge coverage of researchers at different knowledge levels suitable for the training resource. Distribution similarity is calculated for these two features to compare their distribution patterns at different knowledge levels. Similarity calculation methods, such as cosine similarity, can be used to weight the results in combination with the associated edge weights to generate a population suitability index. This population suitability index reflects the degree of match between the researcher's academic background and the target population for the training resource.

[0090] Step S145: Based on the weight value of each associated edge in the scientific research knowledge association structure, a weight coefficient is assigned to the knowledge overlap index, the difficulty connection index, the demand satisfaction index and the population adaptation index.

[0091] Based on the weights of each edge in the scientific research knowledge association structure, corresponding weight coefficients are assigned to the knowledge overlap index, difficulty connection index, demand satisfaction index, and population adaptation index. Edge weights reflect the closeness of the connection between entities, so when assigning weight coefficients, the importance of these connections to the matching results is taken into account. For example, if the edge weight of the mastery relationship between the person entity and the knowledge entity is high, the weight coefficient of the knowledge overlap index may be relatively large.

[0092] Step S146: Multiply the knowledge overlap index, the difficulty connection index, the demand satisfaction index and the population adaptation index by their corresponding weight coefficients and sum them up to generate a matching correlation value between a single resource and a single person.

[0093] The knowledge overlap index, difficulty connection index, demand satisfaction index, and population adaptability index are multiplied by their corresponding weight coefficients, and then these products are added together to obtain the matching correlation value between a single resource and a single person. This matching correlation value comprehensively considers multiple matching factors and can more accurately reflect the degree of match between a researcher and a training resource.

[0094] Step S147: Calculate the matching correlation values ​​of all personnel and all resources to generate a matching correlation matrix containing the matching values ​​of all personnel-resource pairs.

[0095] The matching correlation values ​​described above are calculated for each researcher and each training resource, and the results are organized into a matching correlation matrix. The rows of the matching correlation matrix represent researchers, and the columns represent training resources. Each element in the matching correlation matrix represents the matching correlation value between the corresponding researcher and training resource. This matching correlation matrix provides a comprehensive picture of the matching status between all researchers and training resources.

[0096] Step S150: Generate a scientific research training resource matching result according to the matching association matrix and the dynamic update rule of the scientific research knowledge association structure.

[0097] After obtaining the matching correlation matrix, the final scientific research training resource matching results are generated by combining the dynamic update rules of the scientific research knowledge association structure.

[0098] Step S151: sorting the matching correlation values ​​in the matching correlation matrix in descending order to generate a resource ranking list corresponding to each researcher.

[0099] Each row in the matching correlation matrix (i.e., the matching correlation values ​​of all training resources corresponding to each researcher) is sorted in descending order, and the training resources are arranged from highest to lowest matching correlation with the researcher, generating a ranked resource list for each researcher. This ranked resource list can intuitively demonstrate the order in which each researcher is matched with different training resources.

[0100] Step S152: Filter out resources with matching correlation values ​​greater than a preset threshold from the resource ranking list to form a candidate resource set.

[0101] A preset threshold is set. Training resources with a matching correlation value greater than the threshold are filtered from each researcher's ranked resource list. These resources are then grouped together to form a candidate resource set. This threshold is used to filter out resources with a low matching degree with the researcher, retaining only those with a high matching probability.

[0102] Step S153: Based on the historical learning relationship between the personnel entity and the resource entity in the scientific research knowledge association structure, the number of repeated recommendations of each resource in the candidate resource set is counted. The number of repeated recommendations indicates the frequency with which the resource is selected by people with the same academic background.

[0103] Leveraging the historically learned relationships between person and resource entities within the research knowledge association structure, we count the number of repeated recommendations for each resource in the candidate resource set. Specifically, we analyze how researchers with similar academic backgrounds select these resources and count the number of times each resource is selected. This number of repeated recommendations reflects the resource's popularity and applicability among researchers with similar academic backgrounds.

[0104] Step S154: For the resources in the candidate resource set, adjust their matching association values ​​according to the latest research dynamics of the knowledge entity in the scientific research knowledge association structure, wherein the adjustment process is based on the research popularity of the knowledge entity and the weight value of the resources covering the knowledge entity.

[0105] Taking into account the dynamic changes in the scientific research field, it is necessary to adjust the matching correlation values ​​of resources in the candidate resource set according to the latest research trends of knowledge entities in the scientific research knowledge association structure.

[0106] For example, step S1541: extracting paper data published within a preset time span from the academic paper database, and counting the frequency of occurrence of each knowledge entity in the paper as the research heat value of the knowledge entity.

[0107] We select papers within a preset time span from an academic paper database, perform text analysis on these papers, and count the frequency of occurrence of each knowledge entity in the papers. This frequency serves as the research popularity value of the knowledge entity, reflecting the level of attention it has received in the current scientific research field.

[0108] Step S1542: normalize the research heat value to generate a standardized heat value in the range of 0-1.

[0109] To facilitate subsequent calculations and comparisons, the research interest values ​​are normalized. Using an appropriate normalization method, the research interest values ​​are mapped to a range of 0-1 to obtain standardized interest values. This ensures that the research interest values ​​of different knowledge entities are comparable.

[0110] Step S1543: extract the knowledge entities covered by each resource in the candidate resource set in the scientific research knowledge association structure and their corresponding coverage relationship weight values.

[0111] The knowledge entities covered by each resource in the candidate resource set and the coverage relationship weights between these resources and knowledge entities are extracted from the scientific research knowledge association structure. These weights reflect the coverage and importance of the resources to the knowledge entities.

[0112] Step S1544: Calculate the heat adjustment coefficient of each resource, where the adjustment coefficient is the sum of the product of the normalized heat value of the covered knowledge entity and the covered relationship weight value.

[0113] For each resource, the normalized popularity value of the knowledge entity it covers is multiplied by the corresponding coverage relationship weight value, and then these products are added together to obtain the resource's popularity adjustment coefficient. This popularity adjustment coefficient comprehensively considers the research popularity and coverage level of the knowledge entities covered by the resource.

[0114] Step S1545: multiplying the original matching relevance value of each resource by the heat adjustment coefficient to generate an adjusted matching relevance value.

[0115] Multiply the original matching relevance value of each resource in the candidate resource set by the corresponding heat adjustment coefficient to obtain the adjusted matching relevance value. This allows the matching relevance value to better reflect the applicability of the resource in the current scientific research environment.

[0116] Step S155: Use the adjusted matching correlation value as the final sorting basis to generate a personalized resource matching list for each researcher.

[0117] Based on the adjusted matching relevance values, each researcher's candidate resource set is re-ranked to generate a personalized resource matching list for each researcher. This personalized resource matching list takes into account the researcher's own characteristics, resource matching status, and the latest research trends of the knowledge entity, providing researchers with more accurate training resource recommendations.

[0118] Step S156: combine all personalized resource matching lists into a scientific research and training resource matching result.

[0119] Each researcher's personalized resource matching list is combined to form the final research training resource matching results. This research training resource matching result can provide researchers with training resource recommendations that meet their needs and interests, promoting their learning and development.

[0120] The entire data processing process involves data collection, storage, and use, which may involve privacy-sensitive data, such as the personal information and academic background of researchers. To protect this privacy-sensitive data, a series of privacy protection and anti-leakage technologies are adopted. During the data collection stage, we adhere to the principles of legality, legitimacy, and necessity, only collecting necessary data related to matching scientific research and training resources, and obtaining explicit authorization from the data owner. In terms of data storage, encryption technology is used to encrypt and store data to ensure the security of the data during storage. During data use, the data is anonymized to remove information that can directly identify individuals. At the same time, access control technology is used to restrict access and processing of the data to authorized personnel. In addition, data is backed up and monitored regularly to promptly identify and address possible security vulnerabilities and ensure the security of privacy-sensitive data.

[0121] Figure 2A schematic diagram illustrates exemplary hardware and software components of a knowledge graph-based scientific research and training resource data matching system 100 that can implement the concepts of the present application, as provided in some embodiments of the present application. For example, the processor 120 can be used in the knowledge graph-based scientific research and training resource data matching system 100 to perform the functions of the present application.

[0122] The knowledge graph-based scientific research and training resource data matching system 100 can be a general-purpose server or a special-purpose server, both of which can be used to implement the knowledge graph-based scientific research and training resource data matching method of this application. Although only one server is shown in this application, for convenience, the functions described in this application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0123] For example, the scientific research and training resource data matching system 100 based on the knowledge graph may include a network port 110 connected to the network, one or more processors 120 for executing program instructions, a communication bus 130, and storage media 140 in different forms, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the scientific research and training resource data matching system 100 based on the knowledge graph may also include program instructions stored in ROM, RAM, or other types of non-transitory storage media, or any combination thereof. The method of the present application can be implemented according to these program instructions. The scientific research and training resource data matching system 100 based on the knowledge graph also includes an I / O interface 150 between the computer and other input and output devices.

[0124] For ease of explanation, only one processor is described in the knowledge graph-based scientific research and training resource data matching system 100. However, it should be noted that the knowledge graph-based scientific research and training resource data matching system 100 in this application may also include multiple processors, so the steps performed by one processor described in this application may also be performed jointly or individually by multiple processors. For example, if the processor of the knowledge graph-based scientific research and training resource data matching system 100 executes step A and step B, it should be understood that step A and step B may also be performed jointly by two different processors or individually in one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor execute steps A and B together.

[0125] In addition, an embodiment of the present invention also provides a readable storage medium, in which computer-executable instructions are preset. When the processor executes the computer-executable instructions, the above-mentioned scientific research and training resource data matching method based on the knowledge graph is implemented.

[0126] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, multiple features are sometimes combined into one embodiment, figure or description thereof.

Claims

1. A scientific research and training resource data matching method based on knowledge graph, characterized in that: The method comprises: Constructing a scientific research knowledge association structure, wherein the scientific research knowledge association structure includes entity nodes, association edges, and attribute labels, wherein the entity nodes include personnel nodes, resource nodes, and knowledge nodes; Based on the scientific research knowledge association structure, feature extraction processing is performed on the scientific research personnel data to generate a personnel feature set, wherein the personnel feature set includes academic background features, learning trajectory features, and demand preference features; Based on the scientific research knowledge association structure, feature extraction processing is performed on the training resource data to generate a resource feature set, wherein the resource feature set includes content coverage features, difficulty gradient features, and adapted population features; Performing cross-dimensional matching calculation on the personnel feature set and the resource feature set through the associated edge weights of the scientific research knowledge association structure to generate a matching association matrix; According to the matching correlation matrix and the dynamic update rules of the scientific research knowledge association structure, a scientific research training resource matching result is generated.

2. The scientific research and training resource data matching method based on knowledge graph according to claim 1 is characterized in that: The scientific research knowledge association structure is constructed, and the scientific research knowledge association structure includes entity nodes, association edges and attribute labels. The entity nodes include personnel nodes, resource nodes and knowledge nodes, including: Collect multi-source data from academic paper databases, training course platforms, and scientific research personnel management systems as the original data set; Performing entity recognition processing on the original data set to extract personnel entities, resource entities, and knowledge entities, wherein the personnel entities correspond to individual scientific researchers, the resource entities correspond to specific training courses, and the knowledge entities correspond to subject knowledge points; Performing relationship extraction processing on the original data set to establish inter-entity association edges, wherein the association edges include the mastery relationship between the personnel entity and the knowledge entity, the coverage relationship between the resource entity and the knowledge entity, and the historical learning relationship between the personnel entity and the resource entity; Performing attribute annotation processing on the entity nodes and setting attribute labels, wherein the attribute labels include the research direction of the personnel entity, the teaching form of the resource entity, and the knowledge level of the knowledge entity; Performing weight calibration on the associated edges, and calculating associated edge weight values ​​based on co-occurrence frequencies, wherein the associated edge weight values ​​represent the closeness of the association between entities; The entity nodes, associated edges, attribute labels and weight values ​​are organized into the scientific research knowledge association structure.

3. The scientific research and training resource data matching method based on knowledge graph according to claim 2 is characterized in that: The weight calibration process is performed on the associated edges, and the associated edge weight value is calculated based on the co-occurrence frequency, wherein the associated edge weight value represents the closeness of the association between entities, including: The co-occurrence times of personnel entities and knowledge entities in academic papers are counted as the original weight value of the mastery relationship; Count the co-occurrence times of resource entities and knowledge entities in the course outline as the original weight value of the coverage relationship; Count the co-occurrence times of the personnel entity and the resource entity in the training records as the original weight value of the historical learning relationship; Normalizing the original weight values ​​to generate standardized weight values ​​in the range of 0-1; The normalized weight value is smoothed to generate a final associated edge weight value.

4. The scientific research and training resource data matching method based on knowledge graph according to claim 1 is characterized in that: Based on the scientific research knowledge association structure, feature extraction processing is performed on the scientific researcher data to generate a personnel feature set, which includes academic background features, learning trajectory features, and demand preference features, including: Extract educational background information, completed training records, and training application texts from the scientific researcher data as input data; Based on the mastery relationship between the personnel entity and the knowledge entity in the scientific research knowledge association structure, the educational background information is analyzed for knowledge coverage, and the hierarchical distribution and quantity ratio of the knowledge entities mastered by the scientific researchers are extracted as academic background features; Based on the historical learning relationship between the personnel entity and the resource entity in the scientific research knowledge association structure, a time series analysis is performed on the completed training records to extract the difficulty change trend of the scientific researcher's learning resources and the knowledge coverage expansion path as learning trajectory features; Based on the covering relationship between the knowledge entities and resource entities in the scientific research knowledge association structure, the training application text is semantically parsed to extract the knowledge entities to be learned and the desired resource forms mentioned by the researchers as demand preference features; The academic background features, the learning trajectory features, and the demand preference features are combined into the personnel feature set.

5. The scientific research and training resource data matching method based on knowledge graph according to claim 4 is characterized in that: The historical learning relationship between the personnel entity and the resource entity based on the scientific research knowledge association structure is used to perform time series analysis on the completed training records, and extract the difficulty change trend of the scientific researcher's learning resources and the knowledge coverage expansion path as learning trajectory features, including: Arrange the completed training records of researchers in chronological order to form a learning time series; For each training resource in the learning time series, extract the knowledge level span value of the difficulty gradient feature corresponding to the training resource in the scientific research knowledge association structure; Calculate the difference in knowledge level span values ​​of adjacent training resources to generate a difficulty change rate sequence; Performing sliding window averaging processing on the difficulty change rate sequence to generate a difficulty change trend curve; For each training resource in the learning time series, extracting a knowledge association network of content coverage features corresponding to the training resource in the scientific research knowledge association structure; Through the node expansion analysis of the knowledge association network, the connection path of the knowledge coverage from the initial node to the newly added node in the learning process of researchers is extracted, and the knowledge coverage expansion path is generated; The difficulty change trend curve and the knowledge coverage expansion path are combined into a learning trajectory feature.

6. The scientific research and training resource data matching method based on knowledge graph according to claim 1 is characterized in that: Based on the scientific research knowledge association structure, feature extraction processing is performed on the training resource data to generate a resource feature set, which includes content coverage features, difficulty gradient features, and adapted population features, including: Extract course syllabus text, historical student data and course evaluation records from training resource data as input data; Based on the coverage relationship between resource entities and knowledge entities in the scientific research knowledge association structure, the course outline text is mapped to knowledge nodes, and the hierarchical distribution and association network of the course coverage knowledge entities are extracted as content coverage features; Based on the knowledge level attributes of the knowledge entities in the scientific research knowledge association structure, the order of knowledge nodes in the course outline text is analyzed and processed in increasing difficulty, and the knowledge level span from basic to advanced and the connection strength of adjacent levels are extracted as difficulty gradient features; Based on the historical learning relationship between the personnel entity and the resource entity in the scientific research knowledge association structure, the academic background information in the history student data is statistically analyzed and processed to extract the knowledge coverage level distribution and learning trajectory characteristics of the course history students as the adaptation population characteristics; The content coverage feature, the difficulty gradient feature, and the adapted population feature are combined into the resource feature set.

7. The method for matching scientific research and training resource data based on knowledge graph according to claim 6 is characterized in that: The knowledge level attributes of the knowledge entities based on the scientific research knowledge association structure are analyzed and processed in increasing difficulty for the sequence of knowledge nodes in the course outline text, and the knowledge level span from basic to advanced and the connection strength of adjacent levels are extracted as difficulty gradient features, including: Identify the knowledge entities involved in the course outline text and obtain their knowledge hierarchical attribute values ​​in the scientific research knowledge association structure; Arrange the knowledge entities in the order of the chapters in the syllabus to form a knowledge node sequence; Calculate the difference between the knowledge level attribute values ​​of the first node and the last node in the knowledge node sequence as the knowledge level span; Calculate the difference between the knowledge level attribute values ​​of adjacent nodes in the knowledge node sequence to generate a level-increasing difference sequence; Counting the proportion of the hierarchical increasing difference sequence that is less than or equal to a preset threshold value as the connection strength of adjacent hierarchies; The knowledge level span and the connection strength are combined into a difficulty gradient feature.

8. The method for matching scientific research and training resource data based on knowledge graph according to claim 1 is characterized in that: The cross-dimensional matching calculation is performed on the personnel feature set and the resource feature set through the associated edge weights of the scientific research knowledge association structure to generate a matching association matrix, including: Extracting the knowledge coverage level distribution of the academic background features in the personnel feature set, and performing intersection calculation with the knowledge coverage level distribution of the content coverage features in the resource feature set to generate a knowledge overlap index; Extracting the difficulty change trend of the learning trajectory feature in the personnel feature set, performing trend consistency analysis on the knowledge level span of the difficulty gradient feature in the resource feature set, and generating a difficulty cohesion index; Extracting the knowledge entities to be learned with demand preference characteristics from the personnel feature set, performing path matching processing on the knowledge association network with content coverage characteristics from the resource feature set, and generating a demand satisfaction index; Extract the knowledge coverage ratio of the academic background features in the personnel feature set, calculate the distribution similarity with the knowledge coverage level distribution of the adapted population features in the resource feature set, and generate a population adaptation index; Based on the weight values ​​of the associated edges in the scientific research knowledge association structure, weight coefficients are assigned to the knowledge overlap index, the difficulty connection index, the demand satisfaction index, and the population adaptability index; The knowledge overlap index, the difficulty connection index, the demand satisfaction index and the population adaptation index are multiplied by their corresponding weight coefficients and then summed to generate a matching correlation value between a single resource and a single person; The matching correlation values ​​of all personnel and all resources are calculated to generate a matching correlation matrix containing the matching values ​​of all personnel-resource pairs.

9. The method for matching scientific research and training resource data based on knowledge graph according to claim 1 is characterized in that: Generating a scientific research and training resource matching result according to the matching correlation matrix and the dynamic updating rule of the scientific research knowledge association structure includes: Sorting the matching correlation values ​​in the matching correlation matrix in descending order to generate a resource ranking list corresponding to each scientific researcher; Filtering out resources with matching correlation values ​​greater than a preset threshold from the resource ranking list to form a candidate resource set; Based on the historical learning relationship between the personnel entity and the resource entity in the scientific research knowledge association structure, the number of repeated recommendations of each resource in the candidate resource set is counted. The number of repeated recommendations indicates the frequency with which the resource is selected by people with the same academic background. For the resources in the candidate resource set, their matching relevance values ​​are adjusted according to the latest research trends of the knowledge entities in the scientific research knowledge association structure, wherein the adjustment process is based on the research popularity of the knowledge entity and the weight value of the resource covering the knowledge entity; The adjusted matching correlation value is used as the final ranking basis to generate a personalized resource matching list for each researcher; All personalized resource matching lists are combined into scientific research and training resource matching results.

10. A scientific research and training resource data matching system based on knowledge graph, characterized in that: It includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the scientific research and training resource data matching method based on knowledge graph as described in any one of claims 1 to 9 above.

Images