A Method for Evaluating the Academic Influence of Researchers Based on Semantic Mining and Entropy Weighting

Abstract

This invention discloses a method for evaluating the academic influence of researchers based on semantic mining and entropy weighting. It constructs a unique digital identity for researchers through knowledge graph alignment technology, connecting heterogeneous data from multiple sources; it extracts deep academic features using large-scale model semantic mining and graph neural network analysis; it promotes objective, multi-dimensional evaluation of researchers through a combination of entropy weighting, coupling coordination degree, and TOPSIS; it establishes an infectious disease transmission model and Prophet time-series prediction to achieve dynamic influence characterization; and it creates a self-learning closed-loop system through collaborative optimization and parameter evolution. Finally, it outputs a multi-dimensional profile and interpretable report of researchers, providing scientific decision support for research management.

Description

Technical Field

[0001] This invention relates to a method for evaluating academic influence, and more particularly to a method for evaluating the academic influence of researchers based on semantic mining and entropy weighting, belonging to the field of big data technology. Background Technology

[0002] The evaluation of researchers' academic influence is a crucial basis for the allocation of research resources, talent recruitment, and professional title assessment. The objectivity, accuracy, and timeliness of the evaluation results directly affect the scientific level of research management decisions. With the continuous growth of academic research output and the deepening of interdisciplinary collaboration, emerging research directions are constantly appearing, and the forms of researchers' knowledge contributions are becoming increasingly diverse. Traditional methods of evaluating academic influence are facing unprecedented challenges.

[0003] Currently, the academic community and research management departments generally use linear summation models based on indicators such as the number of projects, funding amounts, and paper output for evaluation. This method fails to reflect the differences in the quality of academic achievements and the depth of knowledge contribution. Simply adding up projects of different levels and types obscures the essential difference in academic leadership between major and general projects. While some evaluation methods attempt to introduce project type scores and regional adjustment coefficients, their score setting relies entirely on human experience, resulting in excessive subjectivity. It is difficult to establish unified evaluation standards among different disciplines and evaluators, leading to severe distortion of evaluation results when comparing across fields.

[0004] At the level of text mining and semantic understanding, existing technologies mostly employ mechanical word segmentation based on dictionaries and LDA topic models for academic text analysis. These methods struggle with emerging and complex concepts such as "digital governance," "the Chinese nation as a community," and "new-quality productivity," as lag in dictionary updates leads to segmentation errors or semantic fragmentation, failing to accurately identify the complete semantics of these terms. LDA topic models represent documents as probability distributions of topics, essentially a shallow semantic modeling based on word frequency statistics, but they struggle to capture deep semantic connections and contextual dependencies between words. Furthermore, existing technologies cannot effectively identify the comprehensive characteristics of interdisciplinary knowledge. When a researcher's findings involve multiple fields, traditional methods cannot quantify the researcher's unique contribution as a "knowledge bridge" across disciplines.

[0005] From a methodological perspective, existing technologies generally suffer from the dual limitations of subjective weighting and static evaluation. Subjective weighting relies on experts' subjective judgments of the importance of each indicator, and the weighting results are significantly influenced by experts' knowledge background and personal preferences, making it difficult to guarantee the stability and repeatability of the evaluation results. On the other hand, the existing annual ranking mechanism adopts a static evaluation paradigm, focusing only on the historical accumulation of researchers at a specific point in time, completely ignoring the temporal evolution of academic influence and the law of knowledge half-life. It cannot depict the dynamic decline of researchers' research enthusiasm, nor can it identify high-potential young researchers who are in the academic rise phase and have great development potential despite their relatively short historical accumulation.

[0006] In summary, existing technologies suffer from several problems: difficulty in associating multi-source heterogeneous data, insufficient semantic mining of emerging composite concepts, strong subjectivity in evaluation indicators, and inability of static evaluation to depict the dynamic evolution of academic influence. Summary of the Invention

[0007] The technical problem to be solved by this invention is to provide a method for evaluating the academic influence of researchers based on semantic mining and entropy weighting, forming a complete closed-loop system from data collection to evaluation output.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0009] A method for evaluating the academic influence of researchers based on semantic mining and entropy weighting includes the following steps: S1. Collect multi-source heterogeneous data and construct a unique digital identity for researchers through entity association technology to form a data lake covering the entire life cycle of "project-results-impact"; S2. Semantically represent the text, automatically identify emerging composite concepts, construct a researcher-topic association network, and quantify the structural position of researchers in the interdisciplinary knowledge system. S3. Extract quantifiable indicators from three dimensions: project quality, academic influence, and social contribution. Automatically determine the weight of each indicator based on the dispersion of the data itself. Examine the balance of researchers' development in each dimension and output a current influence index that can be compared across disciplines. S4. Simulate the diffusion and decay process of academic influence, generate a time-series decay factor, observe historical trajectories to predict future evolution trends, output a potential index, and form a two-dimensional evaluation matrix. S5. Combining current influence, potential index, network structure characteristics, balance coefficient, and time decay factor, the comprehensive weight of each dimension is adaptively optimized through feedback signals to automatically determine the classification boundary and divide the influence level. S6. Generate a multi-dimensional profile including an influence composition radar chart, a development trend chart, and a knowledge network location chart, and output an interpretable report on indicator contribution, confidence interval, and ranking fluctuation analysis. S7. Compare the evaluation results with external benchmarks, optimize the evaluation parameters in reverse based on the deviations, establish a case library and provide prior references for subsequent evaluations, forming a self-evolving closed loop of continuous optimization.

[0010] Further, step S1 specifically includes: 1.1 Multi-source data collection: collecting data from National Social Science Fund projects, academic papers / monographs, and proof of policy adoption; 1.2 Entity Recognition and Candidate Generation: The BiLSTM-CRF model is used to perform named entity recognition on multi-source data to extract the names of researchers, institutions, and collaborators; edit distance, pinyin similarity, and institutional hierarchy are used to generate a candidate matching set for each entity. 1.3 Entity Linking and Disambiguation: Triples are constructed using researchers, institutions, and collaborators. The semantic similarity between candidate entities is calculated using the Siamese-BERT network. Based on this, contextual information is merged, and global disambiguation is completed through a graph matching algorithm. A unique digital identity is assigned to each researcher. 1.4 Data Lake Construction: Using a unique digital identity as the primary key, link project initiation data, subsequent academic achievements, and social impact data to form a data lake covering the entire lifecycle of "project-achievement-impact".

[0011] Further, step S2 specifically includes: 2.1 Text Semantic Vectorization: Cleaning and word segmentation preprocessing are performed on project titles, abstracts, and keywords; the domain-adapted BERT-large model is loaded, and the representation of social science professional terms is learned through the masked language modeling task, transforming each text into a 768-dimensional deep semantic vector, automatically identifying emerging compound concepts and extracting their feature representations; 2.2 Time Series Data Processing: Standardize the format of multi-source time fields and remove outliers; normalize the time series data and generate an initial time series decay factor; 2.3 Construct semantic vectors for topic clustering and build a researcher-topic binary network; then use graph attention network (GAT) for node embedding learning and calculate the network topology index of each researcher as a quantitative feature of academic influence.

[0012] Further, step S3 specifically includes: 3.1 Three-dimensional indicator system: Project quality dimensions: project-level natural ordinal number, funding intensity; Academic outreach dimension: semantic similarity and network structure characteristics between the project theme and subsequent research results; Social contribution dimensions: number of policies adopted, media coverage volume; 3.2. Objective weighting using the entropy weight method: Construct the original index matrix and standardize the positive and negative indicators respectively; Calculate the information entropy of each indicator ,in , Yes, n is Calculate indicator weights , automatically assign higher weights to indicators with greater variability, the weights are determined by the degree of data dispersion, where m is; 3.3 Coupling and Coordination Test: Calculate the comprehensive score of the project based on three dimensions: project quality, academic impact, and social contribution. ; Calculate coupling degree ; Calculate the degree of coordination ,in This is a comprehensive coordination index, where α, β, and γ are the weighting coefficients of the three dimensions mentioned above. Identify researchers with uneven academic performance whose coordination level is below a threshold; 3.4 TOPSIS Comprehensive Evaluation: Constructing an Ideal Solution With negative ideal solution ; Calculate the Euclidean distance between each researcher and the ideal solution and the negative ideal solution. , ; Calculate relative proximity It outputs a current influence index that can be compared across disciplines and years.

[0013] Further, step S4 specifically includes: 4.1 Generation of Time-Series Decay Factor: Due to the time-sensitive nature of academic influence, an exponential decay model is used to calculate the current weight of historical achievements. ,in The decay coefficient is t, which is the number of years between the generation or publication of the academic achievement and the current evaluation time. 4.2 Potential Index Calculation: For application scenarios requiring the identification of high-potential talent, the Prophet time-series model is used to predict the development trend over the next 1-3 years based on historical influence curves, outputting a potential index P. The historical influence curve is composed of the researcher's current influence index over the years. Let the researcher's current influence index be calculated at the [missing information - likely a specific value or index]. The current influence index for the year is Then its historical influence time series is defined as: y={ C 1, C 2,…, Cn},in The number of years for which data is available. This is the current influence index output in step S3. The time series is input into the Prophet time series model for fitting and prediction to obtain the future... Annual forecast Take the prediction period of the first period The projected value for the year serves as a potential index: When the number of data samples is less than T, the system defaults to... , It is the current influence index.

[0014] Further, step S5 specifically includes: 5.1 Factor composition: Current influence index, potential index, network centrality, equilibrium coefficient, and time-series decay factor; 5.2 Adaptive Weight Learning: A Bayesian optimization framework is used to learn the weights of each factor; an initial prior distribution is set, and the objective function is to minimize the error between the evaluation result and the external benchmark. The weight combination is iteratively updated, and the optimal weight vector is output. ; 5.3 Dynamic Threshold Optimization: Based on the actual distribution characteristics of the comprehensive scores, the system automatically determines the boundaries for the high, medium, and low influence levels. An adaptive clustering method based on data distribution is used to divide the score space into three levels, with cluster centers determined by maximizing inter-cluster distance and minimizing intra-cluster distance. When the score distribution shows significant skewness, the system automatically adjusts to a percentile-based level division method to ensure a reasonable distribution of sample sizes for each level. The system records the classification threshold used in each evaluation, providing a reference for subsequent batch evaluations.

[0015] Further, step S6 specifically includes: 6.1 Multi-dimensional profile of researchers: The system generates three types of visualization charts: The influence radar chart shows the score distribution across three dimensions: project quality, academic reach, and social contribution. The development trend chart shows the historical influence curve and the future prediction range; The knowledge network location diagram shows the node positions and connections of researchers in the researcher-topic binary network; 6.2 Interpretability Report: Indicator Contribution Analysis: The marginal contribution of each indicator to the final score is calculated based on the Shapley value. The formula is as follows: Where S is N Any subset of {j}, where j is the index of the indicator whose contribution needs to be calculated, p() is the comprehensive score function calculated using the specified set of indicators, and N is the set of all indicators; Confidence interval: Displays the reliability range of the evaluation results. The confidence interval is generated through Bootstrap resampling. Ranking fluctuation analysis: Compare historical ranking trends to provide dynamic monitoring.

[0016] Further, step S7 specifically includes: 7.1 Result Verification: Compare the system output results with external benchmarks; calculate ranking differences, classification consistency, and indicator deviations; 7.2 Parameter Backward Correction: Based on the comparison error, the following parameters are backward corrected using Bayesian optimization or gradient descent: the selection range of the index of the entropy weight method, the coordination degree threshold of the coupling coordination degree, the parameters of the half-life calculation formula of the dynamic decay model, and the prior distribution of the weights of the multi-factor fusion. 7.3 Case Library Update: Store the optimized parameter combinations in the researcher evaluation case library; label each case with metadata: applicable field, researcher type, and institution type; 7.4 Prior knowledge reuse: When evaluating new researchers, the system automatically queries the case library to match similar researchers; it calls the prior weights of the matched cases as the initial priors for Bayesian optimization; and it forms a continuously optimized, self-iterating evaluation system.

[0017] Compared with existing technologies, this invention has the following advantages and effects: It provides a method for evaluating the academic influence of researchers based on semantic mining and entropy weighting. Addressing the problem of isolated multi-source data, it establishes a unified identity identifier for researchers, connecting data scattered across different sources such as projects, papers, and policies, thus revealing the complete chain of researchers' academic activities. Addressing the difficulty of traditional text mining in capturing emerging concepts and interdisciplinary connections, it constructs a mapping relationship between deep text semantics and academic network structures, quantifying researchers' position in the knowledge system and their interdisciplinary bridging role. Addressing the biases caused by subjective weighting and static evaluation, it establishes an objective evaluation mechanism based on the inherent characteristics of the data, while also introducing temporal factors, ensuring that the evaluation results reflect both current influence and potential for growth and decline. Furthermore, by establishing a comparison and feedback mechanism between the evaluation results and external benchmarks, the system can continuously self-correct, forming a dynamically evolving and constantly improving evaluation system. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method for evaluating the academic influence of researchers based on semantic mining and entropy weighting, which is based on the present invention.

[0019] Figure 2 This is a flowchart of the knowledge graph alignment process of the present invention.

[0020] Figure 3 This is a flowchart of the large-scale semantic mining and academic network analysis process of this invention. Detailed Implementation

[0021] To illustrate in detail the technical solutions adopted by the present invention to achieve the intended technical objectives, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Furthermore, the technical means or technical features in the embodiments of the present invention can be replaced without creative effort. The present invention will be described in detail below with reference to the accompanying drawings and embodiments.

[0022] like Figure 1 As shown, the present invention provides a method for evaluating the academic influence of researchers based on semantic mining and entropy weighting, comprising the following steps: S1. Collect multi-source heterogeneous data and construct a unique digital identity for researchers through entity association technology to form a data lake covering the entire life cycle of "project-results-impact".

[0023] 1.1, such as Figure 2 As shown, multi-source data collection involved crawling project data (approval number, name, principal investigator, institution, funding amount, and level), paper data (title, author, abstract, journal, and citation count), and policy adoption data (document name and adopting institution) from academic data sources such as the National Social Science Fund database, CNKI, and Wanfang Data via data interfaces. To address the missing field for the National Social Science Fund project approval date, the year code was extracted from the project approval number as an estimate of the approval year.

[0024] 1.2 Entity Recognition and Candidate Generation: The BiLSTM-CRF model is used to perform named entity recognition on multi-source data to extract the names of researchers, institutions, and collaborators; edit distance, pinyin similarity, and institutional hierarchy are used to generate a candidate matching set for each entity. 1.3 Entity Linking and Disambiguation: Triples are constructed using researchers, institutions, and collaborators, and the semantic similarity between candidate entities is calculated using the Siamese-BERT network.

[0025] entity pairs Input the BERT-large model and extract the 768-dimensional vector of the [CLS] token. Calculate the cosine similarity:

[0026] Based on this, contextual information is merged to construct a heterogeneous graph. , where nodes Represents entities, edges Relationships are represented. Global disambiguation is performed using a graph matching algorithm, and a unique digital identity is assigned to each researcher.

[0027] 1.4 Data Lake Construction: Using a unique digital identity as the primary key, link project initiation data, subsequent academic achievements, and social impact data to form a data lake covering the entire lifecycle of "project-achievement-impact".

[0028] S2, such as Figure 3 As shown, semantic representation of text is performed, emerging compound concepts are automatically identified, a researcher-topic association network is constructed, and the structural position of researchers in an interdisciplinary knowledge system is quantified.

[0029] 2.1 Text Semantic Vectorization: Cleaning and word segmentation preprocessing are performed on project titles, abstracts, and keywords; a domain-adapted BERT-large model (further pre-trained on a corpus of 5 million social science papers) is loaded, and the representation of social science professional terms is learned through a masked language modeling task. Each text is transformed into a 768-dimensional deep semantic vector, which automatically identifies emerging compound concepts and extracts their feature representations.

[0030] 2.2 Time series data processing: Standardize the format and remove outliers for multi-source time fields (project initiation time, publication time, adoption time); normalize the time series data and generate an initial time series decay factor.

[0031] 2.3. Construct semantic vectors for topic clustering, using the K-Means++ algorithm (K=50) to divide researchers into different research topic clusters. Construct a researcher-topic binary network. ,in For researchers' node set, The set of nodes is a topic node set, and the edge weights represent the proportion of researchers' achievements on that topic. Then, the graph attention network GAT is used to learn node embeddings and calculate the network topology index of each researcher as a quantitative feature of academic influence.

[0032] S3 extracts quantifiable indicators from three dimensions: project quality, academic influence, and social contribution. It automatically determines the weight of each indicator based on the dispersion of the data, examines the balance of researchers' development in each dimension, and outputs a current influence index that can be compared across disciplines.

[0033] 3.1 Three-dimensional indicator system: Project quality dimensions: Natural ordinal number of project level (major = 3, key = 2, general = 1), funding intensity (ten thousand yuan).

[0034] Academic outreach dimension: semantic similarity between project theme and subsequent research results, network structure characteristics (mediation centrality, structural holes).

[0035] Social contribution dimension: number of policies adopted, media coverage volume (number of media reports).

[0036] 3.2. Objective weighting using the entropy weight method: Construct the original index matrix and standardize the positive and negative indicators respectively; Calculate the information entropy of each indicator ,in , Yes, n is Calculate indicator weights It automatically assigns higher weights to indicators with greater variability, determining the weights based on the data dispersion, thus avoiding subjective manual settings. Where m is... 3.3 Coupling and Coordination Test: Calculate the comprehensive score of the project based on three dimensions: project quality, academic impact, and social contribution. ; Calculate coupling degree ; Calculate the degree of coordination ,in This is a comprehensive coordination index, where α, β, and γ are the weighting coefficients of the three dimensions mentioned above. Identify researchers with uneven academic performance whose coordination level is below a threshold; 3.4 TOPSIS Comprehensive Evaluation: Constructing an Ideal Solution (Maximum values ​​of each indicator) and negative ideal solution (Minimum values ​​of each indicator); Calculate the Euclidean distance between each researcher and the ideal solution and the negative ideal solution. , ; Calculate relative proximity It outputs a current influence index that can be compared across disciplines and years.

[0037] S4. Simulate the diffusion and decay process of academic influence, generate a time-series decay factor, observe historical trajectories to predict future evolution trends, output a potential index, and form a two-dimensional evaluation matrix.

[0038] 4.1 Generation of Time-Series Decay Factor: Due to the time-sensitive nature of academic influence, an exponential decay model is used to calculate the current weight of historical achievements. ,in The decay coefficient is t, which is the number of years between the generation or publication of the academic achievement and the current evaluation time.

[0039] 4.2 Potential Index Calculation: For application scenarios requiring the identification of high-potential talent, the Prophet time-series model is used to predict the development trend over the next 1-3 years based on historical influence curves, outputting a potential index P. The historical influence curve is composed of the researcher's current influence index over the years. Let the researcher's current influence index be calculated at the [missing information - likely a specific value or index]. The current influence index for the year is Then its historical influence time series is defined as: y={ C 1, C 2,…, Cn},in The number of years for which data is available. This is the current influence index output in step S3. The time series is input into the Prophet time series model for fitting and prediction to obtain the future... Annual forecast Take the prediction period of the first period The projected value for the year serves as a potential index: When the number of data samples is less than T, the system defaults to... , It is the current influence index.

[0040] S5. Combining current influence, potential index, network structure characteristics, balance coefficient, and time decay factor, the system adaptively optimizes the comprehensive weights of each dimension through feedback signals, automatically determines the classification boundary, and divides the influence level.

[0041] 5.1 Factor composition: Current influence index (from TOPSIS), potential index (from Prophet prediction), network centrality (from graph neural network), balance coefficient (from coupling coordination degree), and time-series decay factor (from infectious disease transmission model).

[0042] 5.2 Adaptive Weight Learning: A Bayesian optimization framework is used to learn the weights of each factor; an initial prior distribution is set (uniform distribution or prior distribution referencing a historical case library); the weight combination is iteratively updated with the objective function of minimizing the error between the evaluation result and the external benchmark, outputting the optimal weight vector. .

[0043] 5.3 Dynamic Threshold Optimization: Based on the actual distribution characteristics of the comprehensive scores, the system automatically determines the boundaries for the high, medium, and low influence levels. An adaptive clustering method based on data distribution is used to divide the score space into three levels, with cluster centers determined by maximizing inter-cluster distance and minimizing intra-cluster distance. When the score distribution shows significant skewness, the system automatically adjusts to a percentile-based level division method to ensure a reasonable distribution of sample sizes for each level. The system records the classification threshold used in each evaluation, providing a reference for subsequent batch evaluations.

[0044] S6 generates a multi-dimensional profile including an influence composition radar chart, a development trend chart, and a knowledge network location chart, and outputs an interpretable report with indicator contribution, confidence interval, and ranking fluctuation analysis.

[0045] 6.1 Multi-dimensional profile of researchers: The system generates three types of visualization charts: The influence radar chart shows the score distribution across three dimensions: project quality, academic reach, and social contribution. The development trend chart shows the historical influence curve and the future prediction range; The knowledge network location diagram shows the node positions and connections of researchers in the researcher-topic binary network.

[0046] 6.2 Interpretability Report: Indicator Contribution Analysis: The marginal contribution of each indicator to the final score is calculated based on the Shapley value. The formula is as follows: Where S is N Any subset of {j}, where j is the index of the indicator whose contribution needs to be calculated, p() is the comprehensive score function calculated using the specified set of indicators, and N is the set of all indicators; Confidence interval: Displays the reliability range of the evaluation results. The confidence interval is generated through Bootstrap resampling. Ranking fluctuation analysis: Compare historical ranking trends to provide dynamic monitoring.

[0047] S7. Compare the evaluation results with external benchmarks, optimize the evaluation parameters in reverse based on the deviations, establish a case library and provide prior references for subsequent evaluations, forming a self-evolving closed loop of continuous optimization.

[0048] 7.1 Result Validation: Compare the system output results with external benchmarks (peer review, subsequent academic performance, authoritative evaluation system); calculate the ranking difference (Kendall tau distance), classification consistency (F1 score), and indicator bias (root mean square error).

[0049] 7.2 Parameter Backward Correction: Based on the comparison error, the following parameters are backward corrected using Bayesian optimization or gradient descent: the selection range of indicators for the entropy weight method (removing indicators with low discrimination), the coordination degree threshold of the coupling coordination degree, the parameters of the half-life calculation formula of the dynamic decay model, and the prior distribution of the weights of the multi-factor fusion.

[0050] 7.3 Case Library Update: Store the optimized parameter combinations in the research personnel evaluation case library; label each case with metadata: applicable field (philosophy / economics / law, etc.), researcher type (young / middle-aged / senior), and institution type (985 / 211 / ordinary university / research institute).

[0051] 7.4 Prior knowledge reuse: When evaluating new researchers, the system automatically queries the case library to match similar researchers (similar in field, researcher type, and institution type); it calls the prior weights of the matched cases as the initial priors for Bayesian optimization; it reduces cold start bias, improves evaluation accuracy, and forms a continuously optimized, self-iterating evaluation system.

[0052] This invention provides a method for evaluating the academic influence of researchers based on semantic mining and entropy weighting. Addressing the problem of isolated multi-source data, it establishes a unified researcher identity identifier, connecting data scattered across different sources such as projects, papers, and policies, thus revealing the complete chain of researchers' academic activities. Addressing the difficulty of traditional text mining in capturing emerging concepts and interdisciplinary connections, it constructs a mapping relationship between deep text semantics and academic network structures, quantifying researchers' positions within the knowledge system and their interdisciplinary bridging role. To address the biases caused by subjective weighting and static evaluation, it establishes an objective evaluation mechanism based on the inherent characteristics of the data, while also introducing temporal factors, ensuring that the evaluation results reflect both current influence and potential for growth and decline. Furthermore, by establishing a comparison and feedback mechanism between the evaluation results and external benchmarks, the system can continuously self-correct, forming a dynamically evolving and constantly improving evaluation system.

[0053] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent substitutions, and improvements made to the above embodiments without departing from the scope of the present invention, based on the technical essence of the present invention and within the spirit and principles of the present invention, shall still fall within the protection scope of the present invention.

Claims

1. A method for evaluating the academic influence of researchers based on semantic mining and entropy weighting, characterized in that... Includes the following steps: S1. Collect multi-source heterogeneous data and construct a unique digital identity for researchers through entity association technology to form a data lake covering the entire life cycle of "project-results-impact"; S2. Semantically represent the text, automatically identify emerging composite concepts, construct a researcher-topic association network, and quantify the structural position of researchers in the interdisciplinary knowledge system. S3. Extract quantifiable indicators from three dimensions: project quality, academic influence, and social contribution. Automatically determine the weight of each indicator based on the dispersion of the data itself. Examine the balance of researchers' development in each dimension and output a current influence index that can be compared across disciplines. S4. Simulate the diffusion and decay process of academic influence, generate a time-series decay factor, observe historical trajectories to predict future evolution trends, output a potential index, and form a two-dimensional evaluation matrix. S5. Combining current influence, potential index, network structure characteristics, balance coefficient, and time decay factor, the comprehensive weight of each dimension is adaptively optimized through feedback signals to automatically determine the classification boundary and divide the influence level. S6. Generate a multi-dimensional profile including an influence composition radar chart, a development trend chart, and a knowledge network location chart, and output an interpretable report on indicator contribution, confidence interval, and ranking fluctuation analysis. S7. Compare the evaluation results with external benchmarks, optimize the evaluation parameters in reverse based on the deviations, establish a case library and provide prior references for subsequent evaluations, forming a self-evolving closed loop of continuous optimization.

2. The method for evaluating the academic influence of researchers based on semantic mining and entropy weighting as described in claim 1, characterized in that: Step S1 specifically involves: 1.1 Multi-source data collection: collecting data from National Social Science Fund projects, academic papers / monographs, and proof of policy adoption; 1.2 Entity Recognition and Candidate Generation: The BiLSTM-CRF model is used to perform named entity recognition on multi-source data to extract the names of researchers, institutions, and collaborators; edit distance, pinyin similarity, and institutional hierarchy are used to generate a candidate matching set for each entity. 1.3 Entity Linking and Disambiguation: Triples are constructed using researchers, institutions, and collaborators. The semantic similarity between candidate entities is calculated using the Siamese-BERT network. Based on this, contextual information is merged, and global disambiguation is completed through a graph matching algorithm. A unique digital identity is assigned to each researcher. 1.4 Data Lake Construction: Using a unique digital identity as the primary key, link project initiation data, subsequent academic achievements, and social impact data to form a data lake covering the entire lifecycle of "project-achievement-impact".

3. The method for evaluating the academic influence of researchers based on semantic mining and entropy weighting as described in claim 1, characterized in that: Step S2 specifically involves: 2.1 Text Semantic Vectorization: Cleaning and word segmentation preprocessing are performed on project titles, abstracts, and keywords; the domain-adapted BERT-large model is loaded, and the representation of social science professional terms is learned through the masked language modeling task, transforming each text into a 768-dimensional deep semantic vector, automatically identifying emerging compound concepts and extracting their feature representations; 2.2 Time Series Data Processing: Standardize the format of multi-source time fields and remove outliers; normalize the time series data and generate an initial time series decay factor; 2.3 Construct semantic vectors for topic clustering and build a researcher-topic binary network; Subsequently, the Graph Attention Network (GAT) was used to learn node embeddings and calculate the network topology index of each researcher as a quantitative feature of academic influence.

4. The method for evaluating the academic influence of researchers based on semantic mining and entropy weighting as described in claim 1, characterized in that: Step S3 specifically involves: 3.1 Three-dimensional indicator system: Project quality dimensions: project-level natural ordinal number, funding intensity; Academic outreach dimension: semantic similarity and network structure characteristics between the project theme and subsequent research results; Social contribution dimensions: number of policies adopted, media coverage volume; 3.

2. Objective weighting using the entropy weight method: Construct the original index matrix and standardize the positive and negative indicators respectively; Calculate the information entropy of each indicator ,in , It is the item in the i-th row and j-th column of the original index matrix, where n is a constant; Calculate indicator weights It automatically assigns higher weights to indicators with greater variability, with the weights determined by the degree of data dispersion, where m is a constant. 3.3 Coupling and Coordination Test: Calculate the comprehensive score of the project based on three dimensions: project quality, academic impact, and social contribution. ; Calculate coupling degree ; Calculate the degree of coordination ,in This is a comprehensive coordination index, where α, β, and γ are the weighting coefficients of the three dimensions mentioned above. Identify researchers with uneven academic performance whose coordination level is below a threshold; 3.4 TOPSIS Comprehensive Evaluation: Constructing an Ideal Solution With negative ideal solution ; Calculate the Euclidean distance between each researcher and the ideal solution and the negative ideal solution. , ; Calculate relative proximity It outputs a current influence index that can be compared across disciplines and years.

5. The method for evaluating the academic influence of researchers based on semantic mining and entropy weighting as described in claim 1, characterized in that: Step S4 specifically involves: 4.1 Generation of Time-Series Decay Factor: Due to the time-sensitive nature of academic influence, an exponential decay model is used to calculate the current weight of historical achievements. ,in The decay coefficient is t, which is the number of years between the generation or publication of the academic achievement and the current evaluation time. 4.2 Potential Index Calculation: For application scenarios requiring the identification of high-potential talent, the Prophet time-series model is used to predict the development trend over the next 1-3 years based on historical influence curves, outputting a potential index P. The historical influence curve is composed of the researcher's current influence index over the years. Let the researcher's current influence index be calculated at the [missing information - likely a specific value or index]. The current influence index for the year is Then its historical influence time series is defined as: y={ C 1, C 2,…, Cn },in The number of years for which data is available. This is the current influence index output in step S3. The time series is input into the Prophet time series model for fitting and prediction to obtain the future... Annual forecast Take the prediction period of the first period The projected value for the year serves as a potential index: When the number of data samples is less than T, the system defaults to... , It is the current influence index.

6. The method for evaluating the academic influence of researchers based on semantic mining and entropy weighting as described in claim 1, characterized in that: Step S5 specifically involves: 5.1 Factor composition: Current influence index, potential index, network centrality, equilibrium coefficient, and time-series decay factor; 5.2 Adaptive Weight Learning: The weights of each factor are learned using a Bayesian optimization framework; We set an initial prior distribution, take minimizing the error between the evaluation result and the external benchmark as the objective function, iteratively update the weight combination, and output the optimal weight vector. ; 5.3 Dynamic Threshold Optimization: Based on the actual distribution characteristics of the comprehensive scores, the system automatically determines the boundaries for the high, medium, and low influence levels. An adaptive clustering method based on data distribution is used to divide the score space into three levels, with cluster centers determined by maximizing inter-cluster distance and minimizing intra-cluster distance. When the score distribution shows significant skewness, the system automatically adjusts to a percentile-based level division method to ensure a reasonable distribution of sample sizes for each level. The system records the classification threshold used in each evaluation, providing a reference for subsequent batch evaluations.

7. The method for evaluating the academic influence of researchers based on semantic mining and entropy weighting as described in claim 1, characterized in that: Step S6 specifically involves: 6.1 Multi-dimensional profile of researchers: The system generates three types of visualization charts: The influence radar chart shows the score distribution across three dimensions: project quality, academic reach, and social contribution. The development trend chart shows the historical influence curve and the future prediction range; The knowledge network location diagram shows the node positions and connections of researchers in the researcher-topic binary network; 6.2 Explanability Report: Indicator Contribution Analysis: The marginal contribution of each indicator to the final score is calculated based on the Shapley value. The formula is as follows: Where S is N Any subset of {j}, where j is the index of the indicator whose contribution needs to be calculated, p() is the comprehensive score function calculated using the specified set of indicators, and N is the set of all indicators; Confidence interval: Displays the reliability range of the evaluation results. The confidence interval is generated through Bootstrap resampling. Ranking fluctuation analysis: Compare historical ranking trends to provide dynamic monitoring.

8. The method for evaluating the academic influence of researchers based on semantic mining and entropy weighting as described in claim 1, characterized in that: Step S7 specifically involves: 7.1 Result Verification: Compare the system output results with external benchmarks; calculate ranking differences, classification consistency, and indicator deviations; 7.2 Parameter Backward Correction: Based on the comparison error, the following parameters are backward corrected using Bayesian optimization or gradient descent: the selection range of the index of the entropy weight method, the coordination degree threshold of the coupling coordination degree, the parameters of the half-life calculation formula of the dynamic decay model, and the prior distribution of the weights of the multi-factor fusion. 7.3 Case Library Update: Store the optimized parameter combinations in the researcher evaluation case library; label each case with metadata: applicable field, researcher type, and institution type; 7.4 Prior knowledge reuse: When evaluating new researchers, the system automatically queries the case library to match similar researchers; it calls the prior weights of the matched cases as the initial priors for Bayesian optimization; and it forms a continuously optimized, self-iterating evaluation system.

Images